CN114886404B - Electronic equipment, device and storage medium - Google Patents

Abstract

The application discloses electronic equipment, a device and a storage medium, and relates to the technical field of medical signal processing. In the application, the waveform characteristics of the rhythm data of the target object in a set time range are extracted to obtain a corresponding original characteristic vector set; then, respectively obtaining a corresponding first feature vector set and a corresponding second feature vector set according to two preset vector element sampling modes; further, fusing the original feature vector set, the first feature vector set and the second feature vector set to obtain a corresponding fused feature vector set, and determining the classification weight of each fused feature vector based on the element correlation degree among the vector elements contained in each fused feature vector in the fused feature vector set; finally, the rhythm type of the rhythm data is determined based on each fusion feature vector and the classification weight corresponding to each fusion feature vector. In this way, the accuracy of classification of the heart rhythm data is improved.

Description

A kind of electronic equipment, device and storage medium

technical field

The present application relates to the technical field of medical signal processing, and in particular to an electronic device, device and storage medium.

Background technique

In recent years, with the continuous improvement of the material level, people pay more and more attention to their own health. Among various diseases, heart disease is not only a relatively common type of disease, but also poses a greater threat to human life and health.

Electrocardiogram (ECG) can objectively reflect the physiological status and working status of various parts of the heart, and is an important means and main basis for diagnosing arrhythmia diseases. However, the identification of ECG still requires experienced medical personnel to accurately diagnose arrhythmia category. Therefore, it is of great practical significance to use intelligent medical equipment and related algorithms to monitor the current patient's heart beating state in time and automatically classify their heart rhythm.

In the prior art, in order to realize the automatic classification of heart rhythm, after obtaining the ECG signal contained in the electrocardiogram, it is normalized to obtain the corresponding heart rhythm data. On this basis, based on the convolutional neural network ( The heart rhythm data classification module constructed by the Convolutional Neural Networks (CNN) model and the encoding-decoding model extracts the characteristics of the heart rhythm data and obtains the corresponding heart rhythm feature information, so as to complete the classification of the heart rhythm data according to the obtained heart rhythm feature information.

However, if the heart rhythm data classification method mentioned above is used to complete the classification of heart rhythm data through the heart rhythm data module, the feature extraction ability of the fusion of the convolutional neural network model and the encoding-decoding model will be insufficient, resulting in inaccurate classification of the heart rhythm data. .

Therefore, with the above method, the accuracy of heart rhythm data classification is low.

Contents of the invention

Embodiments of the present application provide an electronic device, an apparatus, and a storage medium, so as to improve the accuracy of heart rhythm data classification.

In the first aspect, the embodiment of the present application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein, when the processor executes the computer program, Implement the following heart rhythm data classification methods:

Obtain the heart rhythm data of the target object within the set time range, and perform feature extraction on the waveform features of the heart rhythm data to obtain the corresponding original feature vector set;

performing feature compression on each original feature vector included in the original feature vector set according to two preset vector element sampling methods respectively, to obtain corresponding first feature vector set and second feature vector set;

Fusing the original feature vector set, the first feature vector set and the second feature vector set to obtain a corresponding fusion feature vector set;

Based on the set of fusion feature vectors, the element correlation between the vector elements contained in each fusion feature vector is determined to determine the respective classification weights of each fusion feature vector;

A heart rhythm category of the heart rhythm data is determined based on each fusion feature vector and its corresponding classification weight.

In the second aspect, the embodiment of the present application also provides a heart rhythm data classification device, the device comprising:

The obtaining module is used to obtain the heart rhythm data of the target object within the set time range, and perform feature extraction on the waveform features of the heart rhythm data to obtain the corresponding original feature vector set;

The compression module is used to perform feature compression on each original feature vector included in the original feature vector set according to the two preset vector element sampling methods, and obtain the corresponding first feature vector set and second feature vector set;

The fusion module is used to fuse the original feature vector set, the first feature vector set and the second feature vector set to obtain a corresponding fusion feature vector set;

The configuration module is used to determine the respective classification weights of each fusion feature vector based on the fusion feature vector set, the element correlation between the vector elements contained in each fusion feature vector;

An identification module, configured to determine the heart rhythm category of the heart rhythm data based on each fusion feature vector and its corresponding classification weight.

In a possible embodiment, each original feature vector included in the original feature vector set is subjected to feature compression according to the two preset vector element sampling methods to obtain the corresponding first feature vector set and second feature vector When set, the compression module is specifically used for:

Based on two vector element sampling methods, perform global average pooling on each original feature vector to obtain a first semantic information set, and perform global maximum pooling on each original feature vector to obtain a second semantic information set;

Based on the first set of semantic information, a first set of feature vectors is generated, and based on the second set of semantic information, a second set of feature vectors is generated.

In a possible embodiment, when based on the fusion feature vector set, the element correlation between the vector elements contained in each fusion feature vector, the configuration module is specifically used to:

For each fused feature vector, perform the following operations:

Obtain each vector element contained in a fusion feature vector, and perform semantic extraction on the respective element names of each vector element, and obtain the respective semantic information of each element name;

For each element name, do the following:

Comparing the semantic information of an element name with the semantic information of other element names to obtain at least one semantic similarity;

Based on the obtained at least one semantic similarity, respectively determine the vector element corresponding to an element name, and the element correlation between other vector elements.

In a possible embodiment, when determining the respective classification weights of each fusion feature vector based on the element correlation between the vector elements contained in each fusion feature vector in the fusion feature vector set, the configuration module is specifically used to :

For each fused feature vector, perform the following operations:

Respectively obtain each vector element included in a fusion feature vector, and at least one element correlation degree corresponding to each; each element correlation degree represents: the degree of association between the corresponding vector element and one of the other vector elements;

Obtaining respective sub-category weights corresponding to each vector element based on at least one element correlation corresponding to each vector element;

Based on the obtained weights of each sub-category and the respective element values of each vector element, a classification weight of a fused feature vector is obtained.

In a possible embodiment, when determining the heart rhythm category of the heart rhythm data based on each fused feature vector and its corresponding classification weight, the identification module is specifically configured to:

Respectively determine the classification weights corresponding to each fusion feature vector, and the fusion feature weight intervals to which they belong;

The heart rhythm category of the heart rhythm data is determined based on the obtained fusion feature weight intervals and the preset correspondence between the fusion feature weight intervals and heart rhythm categories.

In a third aspect, an electronic device is proposed, which includes a processor and a memory, wherein the memory stores program code, and when the program code is executed by the processor, the processor executes the above-mentioned first The steps of the heart rhythm data classification method described in the aspect.

In a fourth aspect, a computer program product is provided. When the computer program product is invoked by a computer, the computer executes the steps of the heart rhythm data classification method described in the first aspect.

The beneficial effects of this application are as follows:

In the electronic device provided in the embodiment of the present application, the heart rhythm data of the target object within the set time range is obtained, and the waveform features of the heart rhythm data are extracted to obtain the corresponding original feature vector set; then, according to the preset The two vector element sampling methods are used to perform feature compression on each original feature vector contained in the original feature vector set, and obtain the corresponding first feature vector set and second feature vector set; further, the original feature vector set, the first feature vector The vector set and the second feature vector set are fused to obtain a corresponding fused feature vector set, so that based on the fused feature vector set, the element correlation between the vector elements contained in each fused feature vector, the respective classification of each fused feature vector is determined weight; finally, based on each fusion feature vector and its corresponding classification weight, the heart rhythm category of the heart rhythm data is determined.

In this way, the original eigenvector set, the first eigenvector set, and the second eigenvector set are fused to obtain the corresponding fused eigenvector set, so that based on the fused eigenvector set, the vector elements contained in each fused feature vector The correlation between the elements, determine the respective classification weights of each fusion feature vector, and then determine the heart rhythm category of the heart rhythm data based on each fusion feature vector and their corresponding classification weights, avoiding the problem of convolutional neural network model in the prior art The ability to extract features fused with the encoding-decoding model is not enough, which leads to the technical disadvantage of inaccurate classification of heart rhythm data, thus improving the accuracy of heart rhythm data classification.

Furthermore, other features and advantages of the application will be set forth in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the application. The objectives and other advantages of the application may be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.

Description of drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings that need to be used in the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For Those of ordinary skill in the art can also obtain other drawings based on these drawings without any creative effort. In the attached picture:

FIG. 1 exemplarily shows an optional schematic diagram of a system architecture applicable to an embodiment of the present application;

Fig. 2 exemplarily shows a schematic diagram of the overall implementation flow of a heart rhythm data classification method provided by the embodiment of the present application;

Fig. 3 exemplarily shows a method implementation flowchart of a heart rhythm data classification method provided by the embodiment of the present application;

FIG. 4 exemplarily shows a schematic diagram of the composition and structure of a convolutional neural network model provided by an embodiment of the present application;

Fig. 5 exemplarily shows a logical schematic diagram for obtaining the first feature vector set and the second feature vector set provided by the embodiment of the present application;

FIG. 6 exemplarily shows a schematic diagram of the composition and structure of the sequence-to-sequence network model provided by the embodiment of the present application;

FIG. 7 exemplarily shows a schematic diagram of a specific application scenario for obtaining element correlation provided by the embodiment of the present application;

Fig. 8 exemplarily shows a flow chart of implementing a method for determining classification weights of fused feature vectors provided by an embodiment of the present application;

Fig. 9 exemplarily shows a schematic diagram of a specific application scenario for determining the heart rhythm category of the heart rhythm data provided by the embodiment of the present application;

FIG. 10 exemplarily shows a schematic diagram of a specific application scenario based on FIG. 3 provided by the embodiment of the present application;

FIG. 11 exemplarily shows a schematic structural diagram of a heart rhythm data classification device provided by an embodiment of the present application;

FIG. 12 exemplarily shows a schematic structural diagram of an electronic device provided by an embodiment of the present application.

Detailed ways

In order to make the purpose, technical solutions and advantages of the embodiments of the application clearer, the technical solutions of the application will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the application. Obviously, the described embodiments are the Some embodiments of the technical solution, but not all embodiments. Based on the embodiments described in the application documents, all other embodiments obtained by persons of ordinary skill in the art without creative efforts fall within the protection scope of the technical solutions of the present application.

It should be noted that in the description of the present application, "plurality" is understood as "at least two". "And/or" describes the association relationship of associated objects, indicating that there may be three types of relationships, for example, A and/or B may indicate: A exists alone, A and B exist simultaneously, and B exists independently. The connection between A and B can mean: A and B are directly connected and A and B are connected through C. In addition, in the description of the present application, words such as "first" and "second" are only used for the purpose of distinguishing descriptions, and cannot be understood as indicating or implying relative importance, nor can they be understood as indicating or implying order.

In addition, in the technical solution of this application, the collection, dissemination, and use of data, etc., all comply with the requirements of relevant national laws and regulations.

The design idea of the embodiment of the present application is briefly introduced below:

In recent years, cardiovascular disease has become one of the important diseases that threaten people's health, and its morbidity and mortality are increasing year by year. Among them, the occurrence of most cardiovascular diseases is usually accompanied by arrhythmia, that is, arrhythmia is An important cause of heart disease and sudden cardiac death.

In view of the fact that ECG can objectively reflect the physiological status and working status of various parts of the heart, ECG can be used as an important means and main basis for diagnosing arrhythmia diseases. However, the identification of EFC still requires experienced medical personnel to accurately diagnose arrhythmia category. However, even experienced volunteers may still make misjudgment of heart rhythm data. Moreover, due to the huge amount of ECG data and limited medical staff, unavoidable fatigue and other factors will further aggravate the misjudgment of heart rhythm data.

In addition, in order to alleviate the work pressure of medical staff, various related technologies for automatic classification of heart rhythm data have emerged. However, the existing automatic classification technology for heart rhythm data is not capable of extracting the features of heart rhythm data; therefore, there is an urgent need for a A heart rhythm data classification method that improves the feature extraction capability of the heart rhythm data, thereby improving the accuracy of the heart rhythm data classification.

In view of this, an embodiment of the present application provides a heart rhythm data classification method performed by an electronic device, which specifically includes: acquiring the heart rhythm data of a target object within a set time range, and performing feature extraction on the waveform features of the heart rhythm data to obtain The corresponding original feature vector set; then, according to the two preset vector element sampling methods, perform feature compression on each original feature vector contained in the original feature vector set, and obtain the corresponding first feature vector set and second feature vector set ; Further, the original eigenvector set, the first eigenvector set and the second eigenvector set are fused to obtain the corresponding fused eigenvector set, so that based on the fused eigenvector set, the relationship between the vector elements contained in each fused eigenvector Determine the respective classification weights of each fusion feature vector; finally, based on each fusion feature vector and its corresponding classification weight, determine the heart rhythm category of the heart rhythm data; in addition, for the convenience of description and understanding, in the embodiment of this application In this method, the electronic device may be a server executing the heart rhythm data classification method.

In particular, the preferred embodiments of the present application will be described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described here are only used to illustrate and explain the present application, not to limit the present application, and in the absence of conflict , the embodiments of the present application and the features in the embodiments can be combined with each other.

Referring to FIG. 1 , it is a schematic diagram of a system architecture provided by an embodiment of the present application, and the system architecture includes: a server 101 , a terminal device 102 and a data collection device 103 . Server 101 and terminal device 102 can carry out information exchange through communication network, wherein, the communication mode that communication network adopts can include: wireless communication mode and wired communication mode; Terminal device 102 is connected with data collection device 103, and data collection device 103 collects The received heart rhythm data is sent to the terminal device 102.

Exemplarily, the server 101 may access a network through a cellular mobile communication technology, and communicate with the terminal device 102, wherein the cellular mobile communication technology includes, for example, a fifth generation mobile communication (5th Generation MobileNetworks, 5G) technology.

Optionally, the server 101 may access the network through a short-distance wireless communication manner, and communicate with the terminal device 102, wherein the short-distance wireless communication manner, for example, includes a wireless fidelity (Wireless Fidelity, Wi-Fi) technology.

The embodiment of the present application does not impose any restrictions on the number of the above-mentioned devices. As shown in FIG. 1, only the server 101, the terminal device 102 and the data collection device 103 are used as examples for description, and the above-mentioned devices and their respective functions are described below brief introduction.

The server 101 can be an independent physical server, or a server cluster or a distributed system composed of multiple physical servers, and can also provide cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, Cloud servers for basic cloud computing services such as middleware services, domain name services, security services, content delivery network (Content Delivery Network, CDN), and big data and artificial intelligence platforms.

It is worth mentioning that, in the embodiment of the present application, the server 101 is used to obtain the heart rhythm data of the target object within the set time range, and perform feature extraction on the waveform features of the heart rhythm data to obtain the corresponding original feature vector set; then, Perform feature compression on each original feature vector contained in the original feature vector set according to the two preset vector element sampling methods respectively, and obtain the corresponding first feature vector set and second feature vector set; further, the original feature vector set , the first feature vector set and the second feature vector set are fused to obtain the corresponding fused feature vector set, so that based on the fused feature vector set, the element correlation between the vector elements contained in each fused feature vector, determine each fused feature The respective classification weights of the vectors; finally, the heart rhythm category of the heart rhythm data is determined based on each fusion feature vector and its corresponding classification weights.

As an example, refer to FIG. 2, which is an overall implementation flowchart of a heart rhythm data classification method provided by the embodiment of the present application. The server can use the pre-trained convolutional neural network model and sequence-to-sequence network model to Determine the heart rhythm category of the heart rhythm data of the target object acquired within the set time range, wherein the sequence-to-sequence network model can also be called an encoding-decoding model; in addition, in the embodiment of the present application, the heart rhythm data is Electrical beats can be divided into five types of heart rhythms: N (normal or bundle branch block beats), S (abnormal supraventricular beats), V (abnormal ventricular beats), F (fusion beats), and Q ( end can classify beats).

It should be noted that, when the server performs the above-mentioned heart rhythm data classification method steps, it uses a convolutional neural network with channel attention to extract features, and then sends them to a sequence-to-sequence network model with global attention for classification. A dual-attention mechanism is introduced to improve the feature extraction ability of heart rhythm data.

The terminal device 102 is a device that can provide users with voice and/or data connectivity, including: a handheld terminal device with a wireless connection function, a vehicle-mounted terminal device, and the like.

Exemplarily, the terminal device 102 includes, but is not limited to: a mobile phone, a tablet computer, a notebook computer, a palmtop computer, a mobile Internet device (Mobile Internet Device, MID), a wearable device, a virtual reality (Virtual Reality, VR) device, an augmented reality (Augmented Reality, AR) equipment, wireless terminal devices in industrial control, wireless terminal devices in unmanned driving, wireless terminal devices in smart grids, wireless terminal devices in transportation safety, wireless terminal devices in smart cities, or Wireless terminal equipment in smart homes, etc.

In addition, a relevant client may be installed on the terminal device 102, and the client may be software (for example, APP, browser, short video software, etc.), or may be a webpage, a small program, or the like. In the embodiment of the present application, the heart rhythm data can be sent from the terminal device 102 to the server 101 .

The data acquisition device 103 is used to obtain data and record information, and convert the collected signals into analog electrical signals through corresponding sensors, and then convert them into digital signals for storage and preprocessing. Including handheld data collection devices with wireless connectivity, head-mounted data collection devices, and fixed data collection devices.

Exemplarily, the data collection device 103 may be: a batch processing data collector, an industrial data collector, a radio frequency identification (Radio Frequency Identification, RFID) data collector, and other data collection devices (mobile phones, tablet computer, etc.), data acquisition card, etc.

It should be noted that, in the embodiment of the present application, the data acquisition device 103 is described by taking an electrocardiogram as an example, wherein the electrocardiometer can collect the heart rhythm data of the target object in real time, and upload the collected heart rhythm data to the terminal device 102 .

The heart rhythm data classification method provided by the exemplary embodiments of the present application will be described below in conjunction with the above-mentioned system architecture and with reference to the accompanying drawings. The implementation of the application is not limited in this regard.

Referring to Figure 3, it is a flow chart of implementing a heart rhythm data classification method provided in the embodiment of the present application. The execution subject takes the server as an example. The specific implementation process of the method is as follows:

S301: Obtain the heart rhythm data of the target object within a set time range, and perform feature extraction on the waveform features of the heart rhythm data to obtain a corresponding original feature vector set.

Specifically, when step S301 is executed, after the server obtains the heart rhythm data of the target object collected by the data collection device within the set time range, it can analyze the heart rhythm data based on the convolutional neural network model with channel attention. The waveform features are extracted to obtain the corresponding original feature vector set.

Among them, as shown in Figure 4, the convolutional neural network model is mainly composed of two modules, specifically including: CMC module and ML module, the CMC module is used for convolution calculation, extracting local features, and the ML module is used to change the original feature vector set The size of the spatial size stabilizes the training process. It should be noted that the original feature vector set can be presented as the corresponding original feature map.

Furthermore, the CMC module is mainly composed of three parts, namely: convolution, activation function and channel attention. Based on the above composition structure, the CMC module first uses one-dimensional convolution to characterize the waveform characteristics of the acquired heart rhythm data. Extract and obtain the original feature vector set, in which, after each convolution operation, the corresponding activation function is applied to enhance the nonlinear expression ability of the convolutional neural network model, and then complete the subsequent processing of the original feature vector according to the channel attention module The feature compression of each original feature vector contained in the set, that is, to improve the local feature extraction ability.

It should be noted that because the Mish activation function has the advantages of no upper bound, infinite order continuity and smoothness, it can allow information to flow into the convolutional neural network better, and enhance the accuracy and generalization of the convolutional neural network model. Therefore, in the embodiment of the present application, the activation function is the Mish activation function, where the specific expression of the Mish activation function is as follows:

Among them, x represents the input, tanh (*) is a common activation function, and its output mean is 0, therefore, its convergence data is very fast, and the number of iterations can be reduced. The specific expression of the tanh activation function is as follows:

where x represents the input.

In addition, the ML module is mainly composed of two parts, namely: maximum pooling and layer normalization. Based on the composition structure of the ML module, it can avoid the concentration of the original feature vectors processed by the channel attention module. The corresponding fusion feature vector is too long, which increases the calculation amount of the sequence-to-sequence network model and the technical disadvantages of training time. Moreover, while increasing the number of feature channels of the fusion feature vector set, the spatial size is reduced and the convolution neural network is improved. The training efficiency of the network model; in addition, while the spatial size of the fusion feature vector set is changed, the application layer normalization can stabilize the training process and accelerate the convergence speed of the convolutional neural network model.

Obviously, based on the above convolutional neural network model, two downsampling operations are performed on the original feature vector set corresponding to the heart rhythm data.

S302: Perform feature compression on each original feature vector included in the original feature vector set according to two preset vector element sampling methods respectively, to obtain a corresponding first feature vector set and a second feature vector set.

Specifically, as shown in FIG. 5, when executing step S302, after the server obtains the original feature vector set, based on the above two vector element sampling methods, it performs global average pooling on each original feature vector to obtain the first semantic information set, and perform global maximum pooling on each original feature vector to obtain the second semantic information set, thereby generating the first feature vector set based on the obtained first semantic information set, and generating the second semantic information set based on the obtained second semantic information set Two sets of eigenvectors.

，C表示通道数，H表示高度，W表示宽度；接着，分别使用沿通道轴向的全局平均池化 和全局最大池化，集原始特征图中的空间信息，分别产生两种不同的通道语义信息集，即第 一语义信息集

和第二语音信息集

；进一步地，将第一语义信息集

和第二语音信 息集

，分别送入多层感知机中，生成第一通道特征图（第一特征向量集）和第二通道特 征图（第二特征向量集）。 Exemplarily, the server uses the channel attention module in the above convolutional neural network model, firstly, for the input original feature vector set (for example, the original feature map), where the spatial size of the original feature map can be expressed as:

, C represents the number of channels, H represents the height, and W represents the width; then, the global average pooling and global maximum pooling along the channel axis are respectively used to collect the spatial information in the original feature map to generate two different channel semantics information set, the first semantic information set

and the second voice information set

; Further, the first semantic information set

and the second voice information set

, respectively sent to the multi-layer perceptron to generate the first channel feature map (the first feature vector set) and the second channel feature map (the second feature vector set).

Among them, global average pooling: if the input feature map size is H ₁ × W ₁ × C ₁ , average pooling is performed on the features on all channels, and the size of the pooling layer is set to the feature map input size H ₁ × W ₁ , the average element of the entire feature map is output by downsampling, and the output size is 1×1× C ₁ ; global maximum pooling: the input feature map size is H ₂ × W ₂ × C ₂ , and the features on all channels are maximally pooled The size of the pooling layer is set to the feature map input size H ₁ × W ₁ , and the largest element of the entire feature map is output through downsampling, and the output size is 1×1× C ₂ .

It should be noted that since the number of channels of each original feature contained in the above original feature map is not large, in the above multi-layer perceptron, only two fully connected layers with the same number of feature channels are used, and, After each fully connected layer, the Mish activation function is applied to allow the convolutional neural network model to learn more complex nonlinear relationships.

S303: Fusing the original feature vector set, the first feature vector set, and the second feature vector set to obtain a corresponding fused feature vector set.

Specifically, when executing step S303, after obtaining the first feature vector set and the second feature vector set, the server combines the original feature vector set, the first feature vector set, and the second feature vector set based on a preset feature vector fusion method Fusion is performed to obtain the corresponding fusion feature vector set, so that the subsequent sequence-to-sequence network model can perform global feature extraction.

Exemplarily, based on the multi-layer perceptron of the channel attention module in the above-mentioned convolutional neural network model, the server combines the first feature vector set corresponding to the first semantic information set and the second feature vector corresponding to the second semantic information set The vector sets are added, and the Sigmoid activation function is applied to them for normalization to obtain the third feature vector set, and then the third feature vector set is fused with the original feature vector set through the point multiplication operation to obtain the corresponding fusion feature vector set , wherein the above-mentioned specific process of obtaining the third feature vector set and the fusion feature vector set can be expressed as follows:

表示经通道注意力增强得到的第三特征向量 集，

表示Sigmoid激活函数，可将输入重新分布为区间在[0，1]内，在二分类的模型可作 为输出层，其输出结果可表示概率，W 表示多层感知机， F _GAV 表示全局平均池化，F _MAX 表示全 局最大池化，x _c表示原始特征向量集X中的原始特征向量，

表示经通道注意力增强后的融 合特征向量集，上述Sigmoid激活函数的具体表达式如下： Among them, X represents the original feature vector set,

Represents the third feature vector set obtained by channel attention enhancement,

Represents the Sigmoid activation function, which can redistribute the input into the interval [0, 1]. The model in the binary classification can be used as the output layer, and the output result can represent the probability. W represents the multi-layer perceptron, and F _GAV represents the global average pool. , F _MAX represents the global maximum pooling, x _c represents the original feature vector in the original feature vector set X ,

Represents the fused feature vector set enhanced by channel attention, the specific expression of the above Sigmoid activation function is as follows:

where x represents the input.

S304: Based on the element correlation between the vector elements contained in each fusion feature vector in the fusion feature vector set, determine the respective classification weights of each fusion feature vector.

Specifically, when executing step S304, after the server obtains the fusion feature vector set based on the convolutional neural network model, the sequence-to-sequence network model, based on the fusion feature vector set, the vector elements contained in each fusion feature vector The element correlation among them determines the respective classification weights of each fusion feature vector.

Before specifically introducing the method of determining the respective classification weights of each fusion feature vector, a brief description of the sequence-to-sequence network model is given, as shown in Figure 6. The sequence-to-sequence model is also an encoding-decoding model based on a cyclic neural network. Sequence refers to a sequence conversion from sequence A to sequence B. The input sequence is sent to the encoder, and the encoder compresses the input sequence into a vector of a specified length. This process is called encoding; the decoder takes the final state of the encoder as In the initialization state, the previous state is used as input, and the vector sent by the encoder is restored to a sequence. This process is called decoding; finally, the heart rhythm category of the heart rhythm data can be determined based on the output result of the decoder.

It should be noted that the recurrent neural network (Recurrent Neural Network, RNN) is usually used as the basic unit of the encoder and decoder, but in the deep neural network, RNN will be affected by short-term memory, if the sequence length is too long, It is more difficult to send information from an earlier time step to a later time step, causing the final part of the layer to stop updating parameters; in addition, during the backpropagation, RNN is more prone to problems such as gradient disappearance or explosion, so in this paper In the embodiment of the application, a Long Short-Term Memory (LSTM) network (Long Short-Term Memory, LSTM) is used to replace the RNN as the basic unit of the encoder and decoder in the sequence-to-sequence network model.

LSTM is an improved cyclic neural network that can solve the problem that RNN cannot achieve long-distance dependence. Compared with RNN, which has only one transfer state h ^t , LSTM has two transfer states: unit state c ^t , and hidden state h ^t . LSTM removes or adds information to the cell state through the gating structure, where the gate is a way to selectively pass information, including a Sigmoid activation function and point multiplication operation (point-by-point multiplication), LSTM contains three gates Layer: forget gate, input gate, output gate.

Optionally, the encoder and decoder usually process sequence data from left to right, but Bidirectional Long-Short Term Memory (Bi-LSTM) can also be used to make the sequence-to-sequence network model Parameters can be updated from two directions, not only using past information, but also capturing subsequent information; therefore, because Bi-LSTM can better utilize the context information of the data, it usually has better performance than standard LSTM, so in In the embodiment of this application, Bi-LSTM can be used as the basic unit of the encoder and decoder.

Next, based on the above sequence-to-sequence network model, the fusion feature vector set, the element correlation between the vector elements contained in each fusion feature vector can be obtained, as shown in Figure 7, for the above fusion feature vector set, each fusion feature vectors, respectively perform the following operations: obtain each vector element contained in a fusion feature vector, and perform semantic extraction on the respective element names of each vector element, and obtain the respective semantic information of each element name; then, for each element name, respectively Perform the following operations: respectively compare the semantic information of an element name with the semantic information of other element names to obtain at least one semantic similarity, and then determine the correspondence of an element name based on the obtained at least one semantic similarity The vector elements of , and the element correlation between other vector elements.

Exemplarily, taking a fusion feature vector Fus.Feat.V1 as an example, it is assumed that the fusion feature vector Fus.Feat.V1 contains 5 vector elements, that is, the fusion feature vector Fus.Feat.V1 is recorded as: (Vect.Ele.1 , Vect.Ele.2, Vect.Ele.3, Vect.Ele.4, Vect.Ele.5), where each vector element can be represented by the corresponding Key and Value, Key represents the element name of the vector element, Value represents the element value of the vector element. Based on the above-mentioned element correlation acquisition method, the server obtains each vector element and its corresponding element similarity as shown in Table 1:

Table 1

Based on the above table, the server can determine the vector element corresponding to the above element name and the elements between other vector elements according to the semantic information of an element name and the semantic information of other element names and at least one semantic similarity respectively. Correlation, for example, taking the vector element Vect.Ele.1 as an example, the vector element Vect.Ele.1 and the vector element Vect.Ele.2, the vector element Vect.Ele.3, the vector element Vect.Ele.4 and the vector element The semantic similarities between the element names of Vect.Ele.5 are: 88%, 76%, 98%, and 83%. Based on the four semantic similarities obtained above, it can be confirmed that the vector elements Vect.Ele.1 and The element correlations among the vector element Vect.Ele.2, the vector element Vect.Ele.3, the vector element Vect.Ele.4 and the vector element Vect.Ele.5 are: 0.88, 0.76, 0.98, 0.83 in sequence.

Further, after the server obtains the element correlation between the vector elements contained in each fusion feature vector in the fusion feature vector set, based on the fusion feature vector set, the element correlation between the vector elements contained in each fusion feature vector, Determine the respective classification weights of each fusion feature vector. For each fusion feature vector, refer to FIG. 8, which is an implementation flowchart of a method for determining the classification weight of the fusion feature vector provided by the embodiment of the present application. The specific implementation of the method The process is as follows:

S3041: Respectively acquire each vector element included in a fused feature vector, and at least one element correlation degree corresponding to each.

Specifically, when executing step S3041, after the server obtains at least one element correlation degree corresponding to each of all vector elements in the fusion feature vector set based on the above-mentioned method for obtaining element correlation, the server can then use the category of a fusion feature vector Information or identification information, from the element correlation set, obtain each vector element contained in a fusion feature vector, and at least one element correlation corresponding to each; wherein, each element correlation characterizes: the corresponding vector element and other vector elements The degree of association between elements of a vector in .

S3042: Based on at least one element correlation degree corresponding to each vector element, respectively obtain a subcategory weight corresponding to each vector element.

Specifically, when step S3042 is executed, after the server acquires at least one correlation degree of each element corresponding to each vector element, based on at least one correlation degree of each element corresponding to each vector element, respectively obtains the subcategory corresponding to each vector element Weight; among them, the calculation formula of sub-category weight is as follows:

表示向量元素i的子分类权重，

表示向量元素i与第j个向量元素之间的 元素相关度，

表示向量元素

对应的向量元素的个数。 in,

Represents the subcategory weight of vector element i ,

Indicates the element correlation between vector element i and the jth vector element,

Represents vector elements

The number of corresponding vector elements.

，故而，基于上述子分类权重的计算公式，可得 向量元素Vect.Ele.1的子分类权重

。 Exemplarily, still taking the above-mentioned vector element Vect.Ele.1 as an example, the vector element Vect.Ele.2, the vector element Vect.Ele.3, the vector element Vect.Ele.4 and the vector element Vect.Ele.5 The correlation between elements is as follows:

, therefore, based on the calculation formula of the above sub-category weight, the sub-category weight of the vector element Vect.Ele.1 can be obtained

.

S3043: Obtain a classification weight of a fused feature vector based on the obtained weights of each sub-classification and the respective element values of each vector element.

Specifically, when step S3043 is executed, after the server obtains the respective sub-category weights corresponding to each vector element, it can then based on the obtained respective sub-category weights, the respective element values of each vector element, and the classification of the preset fusion feature vector The weight calculation formula is used to obtain the classification weight of the above-mentioned fusion feature vector, wherein the classification weight calculation formula is as follows:

表示上述一个融合特征向量的分类权重，

表示向量元素i的子分类权 重，

表示向量元素i的元素值，m表示上述一个融合特征向量包含的向量元素个数。 in,

Represents the classification weight of the above-mentioned fusion feature vector,

Represents the subcategory weight of vector element i ,

Indicates the element value of vector element i , and m indicates the number of vector elements contained in the above-mentioned fusion feature vector.

Obviously, using the above method effectively avoids the traditional sequence-to-sequence network model, where all content in the sequence enjoys the same importance, but it is not reasonable in actual tasks: first, converting the input into a word vector will already lose part of the information, and in long sequences, the importance of different word vectors is quite different. After introducing the attention mechanism, by re-weighting the probability of the sequence content itself, the weight obtained by attention can be redistributed for each information. Information, that is, the original same intermediate semantics A will be replaced by A _i that changes according to the current content, so that the model itself can learn more important content; in addition, the attention mechanism does not need to be specific to the original sequence-to-sequence network model The structure has been changed, and by adding very few additional parameters and calculations, the ability to extract and fuse global features has been improved, and the accuracy of heart rhythm data classification has been effectively improved.

S305: Determine a heart rhythm category of the heart rhythm data based on each fused feature vector and its corresponding classification weight.

Specifically, as shown in FIG. 9, when step S305 is executed, after the server determines the respective classification weights of each fusion feature vector, it can respectively determine the respective classification weights of each fusion feature vector according to the preset fusion feature weight range , the fusion feature weight intervals to which they belong; then, based on the obtained fusion feature weight intervals and the preset correspondence between the fusion feature weight intervals and heart rhythm categories, the heart rhythm category of the heart rhythm data is determined.

Exemplarily, taking two fusion feature vectors as an example, the weight intervals of each fusion feature and their corresponding heart rhythm categories are shown in Table 2:

Table 2

It should be noted that in the above table, the fused feature vector Fus.Feat.V1 can be divided into two fused feature weight intervals according to the preset fused feature weight interval division rules, and the fused feature vector Fus.Feat.V2 can be divided into two fused feature weight intervals according to the preset The fusion feature weight interval division rule can be divided into three fusion feature weight intervals; further, the server determines the respective classification weights of the fusion feature vector Fus.Feat.V1 and the fusion feature vector Fus.Feat.V2, and the respective belonging After the feature weight interval is fused, the heart rhythm category of the corresponding heart rhythm data can be determined.

Exemplarily, taking the classification weight of the fusion feature vector Fus.Feat.V1 belonging to the fusion feature weight interval 1, and the classification weight of the fusion feature vector Fus.Feat.V2 belonging to the fusion feature weight interval 3 as an example, it can be known that the above heart rhythm data belongs to Type N heart rhythm.

Based on the above heart rhythm data classification method, refer to Figure 10, which is a schematic diagram of a specific application scenario of a heart rhythm data classification method provided by the embodiment of the present application. The server acquires the target object (for example, person A) within the set time range (2022.06.25 15:01:27~2022.06.25 15:01:47) the heart rhythm data Arrhythmia.Data, and perform feature extraction on the waveform feature Wave.Chars of the heart rhythm data Arrhythmia.Data to obtain the corresponding original feature vector Set Orig.Eigen.Set; then, perform feature compression on each original feature vector contained in the original feature vector set Orig.Eig.Set according to the two preset vector element sampling methods (Sampling.Meth1 and Sampling.Meth2), Obtain the corresponding first eigenvector set Fri.Eig.Set and the second eigenvector set Sec.Eig.Set; further, for the original eigenvector set Orig.Eig.Set, the first eigenvector set Fri.Eig.Set and The second feature vector set Sec.Eig.Set is fused to obtain the corresponding fusion feature vector set Fus.Eig.Set, so that based on the fusion feature vector set Fus.Eig.Set, each fusion feature vector (for example, Fus.Feat. V1 and Fus.Feat.V2) The element correlation between the vector elements contained in each (for example, the element correlation Ele.Cor1 and Ele.Cor2 corresponding to the fusion feature vector Fus.Feat.V1, and the fusion feature vector Fus.Feat .V2 corresponding to the element correlation Ele.Cor3 and Ele.Cor4), determine the respective classification weights of each fusion feature vector, in order: 0.85 and 0.76; finally, based on each fusion feature vector (Fus.Feat.V1 and Fus.Feat .V2) and their corresponding classification weights (0.85 and 0.76), determine the heart rhythm category of the heart rhythm data Arrhythmia.Data as N heart rhythm.

To sum up, in the heart rhythm data classification method performed by the electronic device provided in the embodiment of the present application, the heart rhythm data of the target object within the set time range is obtained, and the waveform features of the heart rhythm data are extracted to obtain the corresponding The original feature vector set; then, perform feature compression on each original feature vector included in the original feature vector set according to the two preset vector element sampling methods, and obtain the corresponding first feature vector set and second feature vector set; further Specifically, the original feature vector set, the first feature vector set and the second feature vector set are fused to obtain the corresponding fused feature vector set, so that based on the fused feature vector set, the elements between the vector elements contained in each fused feature vector The degree of correlation determines the respective classification weights of the fusion feature vectors; finally, the heart rhythm category of the heart rhythm data is determined based on the fusion feature vectors and their corresponding classification weights.

In this way, the original eigenvector set, the first eigenvector set, and the second eigenvector set are fused to obtain the corresponding fused eigenvector set, so that based on the fused eigenvector set, the vector elements contained in each fused feature vector The correlation between the elements, determine the respective classification weights of each fusion feature vector, and then determine the heart rhythm category of the heart rhythm data based on each fusion feature vector and their corresponding classification weights, avoiding the problem of convolutional neural network model in the prior art The ability to extract features fused with the encoding-decoding model is not enough, which leads to the technical disadvantage of inaccurate classification of heart rhythm data, thus improving the accuracy of heart rhythm data classification.

It should be noted that, according to the heart rhythm data classification method provided by the embodiment of the present application, the accuracy of the heart rhythm category of the heart rhythm data determined by the server is obtained according to the accuracy, sensitivity, specificity and precision of the recognition result of the heart rhythm category , that is, the heart rhythm data classification method provided in the embodiment of the present application provides the accuracy, sensitivity, specificity and precision of the heart rhythm data classification.

Further, based on the same technical idea, the embodiment of the present application also provides a cardiac rhythm data classification device, which is used to implement the above-mentioned flow of the cardiac rhythm data classification method in the embodiment of the present application. Referring to Figure 11, the heart rhythm data classification device includes: an acquisition module 1101, a compression module 1102, a fusion module 1103, a configuration module 1104, and an identification module 1105, wherein:

The obtaining module 1101 is used to obtain the heart rhythm data of the target object within the set time range, and perform feature extraction on the waveform features of the heart rhythm data to obtain the corresponding original feature vector set;

The compression module 1102 is configured to perform feature compression on each original feature vector included in the original feature vector set according to two preset vector element sampling methods, to obtain corresponding first feature vector set and second feature vector set;

A fusion module 1103, configured to fuse the original feature vector set, the first feature vector set and the second feature vector set to obtain a corresponding fusion feature vector set;

The configuration module 1104 is used to determine the respective classification weights of each fusion feature vector based on the element correlation between the vector elements contained in each fusion feature vector in the fusion feature vector set;

The recognition module 1105 is configured to determine the heart rhythm category of the heart rhythm data based on each fusion feature vector and its corresponding classification weight.

In a possible embodiment, each original feature vector included in the original feature vector set is subjected to feature compression according to the two preset vector element sampling methods to obtain the corresponding first feature vector set and second feature vector When set, the compression module 1102 is specifically used for:

Based on two vector element sampling methods, perform global average pooling on each original feature vector to obtain a first semantic information set, and perform global maximum pooling on each original feature vector to obtain a second semantic information set;

Based on the first set of semantic information, a first set of feature vectors is generated, and based on the second set of semantic information, a second set of feature vectors is generated.

In a possible embodiment, when based on the fusion feature vector set, the element correlation between the vector elements contained in each fusion feature vector, the configuration module 1104 is specifically configured to:

For each fused feature vector, perform the following operations:

Obtain each vector element contained in a fusion feature vector, and perform semantic extraction on the respective element names of each vector element, and obtain the respective semantic information of each element name;

For each element name, do the following:

Comparing the semantic information of an element name with the semantic information of other element names to obtain at least one semantic similarity;

Based on the obtained at least one semantic similarity, respectively determine the vector element corresponding to an element name, and the element correlation between other vector elements.

In a possible embodiment, when determining the respective classification weights of each fusion feature vector based on the element correlation between the vector elements contained in each fusion feature vector in the fusion feature vector set, the configuration module 1104 specifically uses At:

For each fused feature vector, perform the following operations:

Respectively obtain each vector element included in a fusion feature vector, and at least one element correlation degree corresponding to each; each element correlation degree represents: the degree of association between the corresponding vector element and one of the other vector elements;

Obtaining respective sub-category weights corresponding to each vector element based on at least one element correlation corresponding to each vector element;

Based on the obtained weights of each sub-category and the respective element values of each vector element, a classification weight of a fused feature vector is obtained.

In a possible embodiment, when determining the heart rhythm category of the heart rhythm data based on each fused feature vector and its corresponding classification weight, the identification module 1105 is specifically configured to:

Respectively determine the classification weights corresponding to each fusion feature vector, and the fusion feature weight intervals to which they belong;

The heart rhythm category of the heart rhythm data is determined based on the obtained fusion feature weight intervals and the preset correspondence between the fusion feature weight intervals and heart rhythm categories.

Based on the same technical concept, the embodiment of the present application also provides an electronic device, which can implement the flow of the heart rhythm data classification method provided in the above-mentioned embodiments of the present application. In an embodiment, the electronic device may be a server, or a terminal device or other electronic devices. As shown in Figure 12, the electronic equipment may include:

At least one processor 1201, and a memory 1202 connected to at least one processor 1201. The embodiment of the present application does not limit the specific connection medium between the processor 1201 and the memory 1202. In FIG. Take the connection through the bus 1200 as an example. The bus 1200 is represented by a thick line in FIG. 12 , and the connection manners between other components are only for schematic illustration and are not limited thereto. The bus 1200 can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in FIG. 12 , but it does not mean that there is only one bus or one type of bus. Alternatively, the processor 1201 may also be called a controller, and the name is not limited.

In the embodiment of the present application, the memory 1202 stores instructions that can be executed by at least one processor 1201, and at least one processor 1201 executes the instructions stored in the memory 1202 to implement a heart rhythm data classification method discussed above. The processor 1201 may implement functions of various modules in the apparatus shown in FIG. 11 .

Wherein, the processor 1201 is the control center of the device, and various interfaces and lines can be used to connect various parts of the entire control device, and by running or executing instructions stored in the memory 1202 and calling data stored in the memory 1202, the Various functions and processing data of the device, so as to monitor the device as a whole.

In a possible design, the processor 1201 may include one or more processing units, and the processor 1201 may integrate an application processor and a modem processor, wherein the application processor mainly processes the operating system, user interface and application programs etc., the modem processor mainly handles wireless communication. It can be understood that the foregoing modem processor may not be integrated into the processor 1201 . In some embodiments, the processor 1201 and the memory 1202 can be implemented on the same chip, and in some embodiments, they can also be implemented on independent chips.

The processor 1201 can be a general-purpose processor, such as a CPU, a digital signal processor, an application-specific integrated circuit, a field programmable gate array or other programmable logic devices, a discrete gate or transistor logic device, and a discrete hardware component, which can implement or execute the present application. Various methods, steps and logic block diagrams disclosed in the embodiments. A general purpose processor may be a microprocessor or any conventional processor or the like. The steps of a heart rhythm data classification method disclosed in the embodiments of the present application may be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules in the processor.

The memory 1202, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs and modules. The memory 1202 may include at least one type of storage medium, for example, may include flash memory, hard disk, multimedia card, card memory, random access memory (Random Access Memory, RAM), static random access memory (Static Random Access Memory, SRAM), Programmable Read Only Memory (PROM), Read Only Memory (ROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Magnetic Memory, Disk, discs and more. Memory 1202 is any other medium that can be used to carry or store desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory 1202 in this embodiment of the present application may also be a circuit or any other device capable of implementing a storage function, and is used for storing program instructions and/or data.

By designing and programming the processor 1201, the code corresponding to a heart rhythm data classification method introduced in the foregoing embodiments can be solidified into the chip, so that the chip can execute one of the embodiments shown in Figure 3 during operation. Steps of a heart rhythm data classification method. How to design and program the processor 1201 is well known to those skilled in the art and will not be repeated here.

Based on the same inventive concept, the embodiment of the present application also provides a storage medium, the storage medium stores computer instructions, and when the computer instructions are run on the computer, the computer executes the heart rhythm data classification method discussed above.

In some possible implementations, various aspects of the heart rhythm data classification method provided by the present application can also be implemented in the form of a program product, which includes program code. When the program product runs on the device, the program code is used to use The control device executes the steps in a method for classifying cardiac rhythm data according to various exemplary embodiments of the present application described above in this specification.

It should be noted that although several units or subunits of the apparatus are mentioned in the above detailed description, this division is only exemplary and not mandatory. Actually, according to the embodiment of the present application, the features and functions of two or more units described above may be embodied in one unit. Conversely, the features and functions of one unit described above may be further divided to be embodied by a plurality of units.

In addition, while operations of the methods of the present application are depicted in the figures in a particular order, there is no requirement or implication that these operations must be performed in that particular order, or that all illustrated operations must be performed to achieve desirable results. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step for execution, and/or one step may be decomposed into multiple steps for execution.

As will be appreciated by those skilled in the art, the embodiments of the present application may be provided as a method, a system, or a computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

The present application is described with reference to flowcharts and/or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present application. It will be understood that each process and/or block in the flowchart illustrations and/or block diagrams, and combinations of processes and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a general purpose computer, a special purpose computer, an embedded processor, or a processor of other programmable data processing equipment to produce a server such that the instructions executed by the computer or the processor of other programmable data processing equipment produce a server An apparatus for realizing the functions specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.

Any combination of one or more programming languages can be used to write the program code for performing the operation of this application. The programming language includes object-oriented programming languages, such as Java, C++, etc., and also includes conventional procedural programming A language, such as "C" or a similar programming language. The program code may execute entirely on the user computing device, partly on the user device, as a stand-alone software package, partly on the user computing device and partly on a remote computing device, or entirely on the remote computing device or server to execute.

In cases involving a remote computing device, the remote computing device may be connected to the user computing device through any kind of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., using an Internet service provider to connect via the Internet).

These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing apparatus to operate in a specific manner, such that the instructions stored in the computer-readable memory produce an article of manufacture comprising instruction means, the instructions The device realizes the function specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.

These computer program instructions can also be loaded onto a computer or other programmable data processing device, causing a series of operational steps to be performed on the computer or other programmable device to produce a computer-implemented process, thereby The instructions provide steps for implementing the functions specified in the flow chart or blocks of the flowchart and/or the block or blocks of the block diagrams.

Obviously, those skilled in the art can make various changes and modifications to the application without departing from the spirit and scope of the application. In this way, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalent technologies, the present application is also intended to include these modifications and variations.

Claims (11)

1. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor, when executing the computer program, implements a method for classifying cardiac rhythm data comprising:

acquiring rhythm data of a target object in a set time range, and performing feature extraction on waveform features of the rhythm data to obtain a corresponding original feature vector set;

respectively performing feature compression on each original feature vector contained in the original feature vector set according to two preset vector element sampling modes to obtain a corresponding first feature vector set and a corresponding second feature vector set;

fusing the original feature vector set, the first feature vector set and the second feature vector set to obtain corresponding fused feature vector sets;

determining respective classification weights of the fusion feature vectors based on element correlation degrees between vector elements contained in the fusion feature vector set and the fusion feature vectors respectively;

and determining the rhythm type of the rhythm data based on the fusion characteristic vectors and the classification weights corresponding to the fusion characteristic vectors.

2. The electronic device according to claim 1, wherein the performing feature compression on each original feature vector included in the original feature vector set according to two preset vector element sampling manners to obtain a corresponding first feature vector set and a second feature vector set comprises:

based on the two vector element sampling modes, performing global average pooling on each original feature vector to obtain a first semantic information set, and performing global maximum pooling on each original feature vector to obtain a second semantic information set;

generating the first set of feature vectors based on the first set of semantic information, and generating the second set of feature vectors based on the second set of semantic information.

3. The electronic device of claim 1, wherein the element correlation between vector elements included in each of the fused feature vectors based on the set of fused feature vectors comprises:

for each fused feature vector, respectively performing the following operations:

acquiring each vector element contained in one fused feature vector, and performing semantic extraction on the element name of each vector element respectively to obtain semantic information of each element name;

for each element name, respectively performing the following operations:

respectively comparing semantic similarity of semantic information of one element name with semantic information of other element names to obtain at least one semantic similarity;

and respectively determining the element correlation degree between the vector element corresponding to the element name and other vector elements based on the obtained at least one semantic similarity.

4. The electronic device of any of claims 1-3, wherein determining the classification weight for each respective fused feature vector in the set of fused feature vectors based on an element correlation between vector elements included in the respective fused feature vector comprises:

for each fused feature vector, respectively performing the following operations:

respectively obtaining each vector element contained in one fusion feature vector and at least one element correlation degree corresponding to each vector element; each element relevance characterizes: a degree of association between the respective vector element and one of the other vector elements;

respectively obtaining sub-classification weights respectively corresponding to the vector elements based on at least one element correlation degree respectively corresponding to the vector elements;

and obtaining the classification weight of the fusion feature vector based on the obtained sub-classification weights and the respective element values of the vector elements.

5. The electronic device of any one of claims 1-3, wherein the determining a rhythm class of the rhythm data based on the respective fused feature vectors and their respective corresponding classification weights comprises:

respectively determining the classification weight corresponding to each fusion feature vector and the fusion feature weight interval to which each fusion feature vector belongs;

and determining the rhythm type of the rhythm data based on the obtained fusion characteristic weight intervals and the corresponding relation between the preset fusion characteristic weight intervals and the rhythm type.

6. A cardiac rhythm data classification apparatus, comprising:

the acquisition module is used for acquiring the heart rhythm data of a target object within a set time range, and extracting the characteristics of the waveform characteristics of the heart rhythm data to obtain a corresponding original characteristic vector set;

the compression module is used for performing feature compression on each original feature vector contained in the original feature vector set according to two preset vector element sampling modes to obtain a corresponding first feature vector set and a corresponding second feature vector set;

the fusion module is used for fusing the original feature vector set, the first feature vector set and the second feature vector set to obtain a corresponding fusion feature vector set;

the configuration module is used for determining the classification weight of each fused feature vector based on the element correlation degree between the vector elements contained in each fused feature vector in the fused feature vector set;

and the identification module is used for determining the rhythm type of the rhythm data based on each fusion feature vector and the corresponding classification weight.

7. The apparatus according to claim 6, wherein when the feature compression is performed on each original feature vector included in the original feature vector set according to two preset vector element sampling manners, respectively, to obtain a corresponding first feature vector set and a corresponding second feature vector set, the compression module is specifically configured to:

based on the two vector element sampling modes, performing global average pooling on each original feature vector to obtain a first semantic information set, and performing global maximum pooling on each original feature vector to obtain a second semantic information set;

generating the first feature vector set based on the first semantic information set, and generating the second feature vector set based on the second semantic information set.

8. The apparatus according to claim 6, wherein, in said based on the element correlation between the vector elements included in each fused feature vector in the fused feature vector set, the configuration module is specifically configured to:

for each fused feature vector, respectively performing the following operations:

acquiring each vector element contained in one fused feature vector, and performing semantic extraction on the element name of each vector element respectively to obtain semantic information of each element name;

for each element name, the following operations are respectively executed:

respectively comparing semantic similarity of semantic information of one element name with semantic information of other element names to obtain at least one semantic similarity;

and respectively determining the element correlation degree between the vector element corresponding to the element name and other vector elements based on the obtained at least one semantic similarity.

9. The apparatus according to any of claims 6-8, wherein, in said determining the classification weight of each respective fused feature vector based on the element correlation between the vector elements included in each respective fused feature vector in the set of fused feature vectors, the configuration module is specifically configured to:

for each fused feature vector, respectively performing the following operations:

respectively obtaining each vector element contained in one fusion feature vector and at least one element correlation degree corresponding to each vector element; each element relevance characterizes: a degree of association between the respective vector element and one of the other vector elements;

respectively obtaining sub-classification weights corresponding to the vector elements based on at least one element correlation degree corresponding to the vector elements;

and obtaining the classification weight of the fusion feature vector based on the obtained sub-classification weights and the respective element values of the vector elements.

10. The apparatus of any one of claims 6-8, wherein, in the determining the rhythm category of the rhythm data based on the respective fused feature vectors and their respective corresponding classification weights, the identification module is specifically configured to:

respectively determining the classification weight corresponding to each fusion feature vector and the fusion feature weight interval to which each fusion feature vector belongs;

and determining the rhythm type of the rhythm data based on the obtained fusion characteristic weight intervals and the corresponding relation between the preset fusion characteristic weight intervals and the rhythm type.

11. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the electronic device according to any one of claims 1-5.

Images