CN111079841A - Training method and device for target recognition, computer equipment and storage medium - Google Patents
Training method and device for target recognition, computer equipment and storage medium Download PDFInfo
- Publication number
- CN111079841A CN111079841A CN201911300954.3A CN201911300954A CN111079841A CN 111079841 A CN111079841 A CN 111079841A CN 201911300954 A CN201911300954 A CN 201911300954A CN 111079841 A CN111079841 A CN 111079841A
- Authority
- CN
- China
- Prior art keywords
- sample image
- image set
- sample
- categories
- training
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Cheminformatics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Image Analysis (AREA)
Abstract
The application relates to a training method and device for target recognition, computer equipment and a storage medium. The method comprises the following steps: acquiring video stream data, wherein the video stream data comprises a plurality of frames of images; reading the image, detecting a corresponding target in the image, and generating a sample image set of a plurality of categories by using the target; training the sample image sets of multiple categories by using an identification model to obtain identification probabilities corresponding to the categories; adjusting the samples in the corresponding sample image set according to the identification probability; and optimizing the recognition model by using the adjusted sample image set, and training the optimized recognition model by using the adjusted sample image set. By adopting the method, the problem of poor model learning effect caused by unbalanced samples can be solved, so that the trained recognition model can carry out accurate target recognition, and the problems of under-fitting and over-fitting of the training model are avoided.
Description
Technical Field
The present application relates to the field of computer technologies, and in particular, to a training method and apparatus for target recognition, a computer device, and a storage medium.
Background
With the development of machine learning and deep learning techniques, more and more users use deep learning techniques to identify targets. In the field of artificial intelligence, machine learning and deep learning need to improve the accuracy of machine judgment by learning a large number of sample cases. Particularly, in the specific target recognition in the field of machine vision, sample images with targets need to be marked, and a large amount of training is performed on a target sample image set, so that a training model has the capability of recognizing multiple target samples simultaneously.
However, in the current training mode of target recognition, data of random entries are randomly extracted and copied, and then the randomly extracted data are added into an original sample, so that the problem of unbalanced learning samples is difficult to solve, and the problems of under-fitting and over-fitting of a training model are easily caused, so that the trained recognition model is difficult to perform accurate target recognition.
Disclosure of Invention
In view of the above, it is necessary to provide a training method, an apparatus, a computer device, and a storage medium for target recognition that can solve the problem of learning sample imbalance.
A training method of object recognition, the method comprising:
acquiring video stream data, wherein the video stream data comprises a plurality of frames of images;
reading the image, detecting a corresponding target in the image, and generating a sample image set of a plurality of categories by using the target;
training the sample image sets of multiple categories by using an identification model to obtain identification probabilities corresponding to the categories;
adjusting the samples in the corresponding sample image set according to the identification probability;
and optimizing the recognition model by using the adjusted sample image set, and training the optimized recognition model by using the adjusted sample image set.
In one embodiment, the training of the sample image set of multiple classes using a recognition model includes:
acquiring the number of samples corresponding to the sample image set of each category;
selecting the minimum sample number as a standard unit amount;
equally dividing the number of samples of the sample image set of each category according to the standard unit amount to obtain a plurality of corresponding standard unit amount sample image sets;
and training the standard unit quantity sample image set corresponding to the sample image set of each category.
In one embodiment, the adjusting the samples in the corresponding sample image set according to the recognition probability includes:
acquiring a sample proportion corresponding to the sample image set;
when the sample proportion reaches a first threshold value, marking the sample image set of the category corresponding to the sample proportion as a sample image set of a majority category;
acquiring target coordinate accuracy corresponding to the sample image sets of the multiple categories;
similarity matching is carried out on the target coordinate accuracy and preset target coordinate accuracy to obtain recognition probabilities corresponding to most categories;
and when the identification probability is greater than the preset identification probability, performing clipping processing on the sample image sets of the majority categories corresponding to the identification probability until the number of samples corresponding to the sample image sets of the majority categories reaches a data balance reasonable range.
In one embodiment, the adjusting the samples in the corresponding sample image set according to the recognition probability includes:
acquiring a sample proportion corresponding to the sample image set;
when the sample proportion reaches a second threshold value, marking the sample image set of the class corresponding to the sample proportion as the sample image set of the minority class;
acquiring target coordinate accuracy corresponding to the sample image sets of the minority categories;
similarity matching is carried out on the target coordinate accuracy and preset target coordinate accuracy, and recognition probabilities corresponding to a few categories are obtained;
and when the identification probability is smaller than the preset identification probability, performing enhancement processing on the sample image sets of the minority categories corresponding to the identification probability until the number of samples corresponding to the sample image sets of the minority categories reaches a data balance reasonable range.
In one embodiment, the enhancing the sample image set of the minority class corresponding to the recognition probability includes:
randomly extracting the sample images in the sample image set of the minority class to obtain original randomly extracted sample images;
randomly adjusting the brightness and the contrast of the original sample image to obtain a newly added sample image;
and carrying out shielding processing on the original sample image to obtain a newly added sample image.
In one embodiment, the optimizing the recognition model by using the adjusted sample image set includes:
and optimizing the recognition model by adjusting a loss function in the recognition model and adding the adjusted sample quantity proportional weight of the sample image set to the recognition model.
An apparatus for training of object recognition, the apparatus comprising:
the acquisition module is used for acquiring video stream data, and the video stream data comprises a plurality of frames of images;
the generating module is used for reading the image, detecting a corresponding target in the image, and generating a sample image set of a plurality of categories by using the target;
the training module is used for training the sample image sets of multiple categories by using a recognition model to obtain recognition probabilities corresponding to the categories;
the adjusting module is used for adjusting the samples in the corresponding sample image set according to the identification probability;
and the optimization module is used for optimizing the recognition model by utilizing the adjusted sample image set and training the optimized recognition model through the adjusted sample image set.
A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the following steps when executing the computer program:
acquiring video stream data, wherein the video stream data comprises a plurality of frames of images;
reading the image, detecting a corresponding target in the image, and generating a sample image set of a plurality of categories by using the target;
training the sample image sets of multiple categories by using an identification model to obtain identification probabilities corresponding to the categories;
adjusting the samples in the corresponding sample image set according to the identification probability;
and optimizing the recognition model by using the adjusted sample image set, and training the optimized recognition model by using the adjusted sample image set.
A computer-readable storage medium, on which a computer program is stored which, when executed by a processor, carries out the steps of:
acquiring video stream data, wherein the video stream data comprises a plurality of frames of images;
reading the image, detecting a corresponding target in the image, and generating a sample image set of a plurality of categories by using the target;
training the sample image sets of multiple categories by using an identification model to obtain identification probabilities corresponding to the categories;
adjusting the samples in the corresponding sample image set according to the identification probability;
and optimizing the recognition model by using the adjusted sample image set, and training the optimized recognition model by using the adjusted sample image set.
The training method, the training device, the computer equipment and the storage medium for target identification are used for acquiring video stream data, wherein the video stream data comprises a plurality of frames of images. Reading the image, detecting a corresponding target in the image, and generating a sample image set of a plurality of categories by using the target. And training the sample image sets of the multiple categories by using the recognition model to obtain recognition probabilities corresponding to the categories. And adjusting the samples in the corresponding sample image set according to the recognition probability. And optimizing the recognition model by using the adjusted sample image set, and training the optimized recognition model by using the adjusted sample image set. Compared with the traditional target recognition training mode, the method has the advantages that the number relation of samples among different types of sample image sets is analyzed, a small number of types of samples is enhanced, a large number of types of samples are weakened, efficient and organic fusion application is achieved, the effect of balancing the number of samples among the types in the sample library is achieved, meanwhile, the learning training quality of the neural network is not affected, namely, the problem of poor model learning effect caused by unbalanced samples is solved by adopting a dynamic processing scheme according to the number relation of the samples among the different types of sample image sets in the sample library, therefore, the trained recognition model can carry out accurate target recognition, and the problems of under-fitting and over-fitting of the training model are avoided.
Drawings
FIG. 1 is a diagram illustrating an exemplary embodiment of a training method for object recognition;
FIG. 2 is a schematic flow chart diagram illustrating a training method for object recognition in one embodiment;
FIG. 3 is a flowchart illustrating the training steps performed on a sample image set of multiple classes using a recognition model in one embodiment;
FIG. 4 is a flow diagram illustrating the steps of the process of conditioning samples in a corresponding sample image set based on recognition probabilities in one embodiment;
FIG. 5 is a flowchart illustrating the steps of the adjustment process for the samples in the corresponding sample image set according to the recognition probability in another embodiment;
FIG. 6 is a block diagram of an embodiment of an apparatus for training object recognition;
FIG. 7 is a diagram illustrating an internal structure of a computer device according to an embodiment.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more apparent, the present application is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.
The training method for target recognition provided by the application can be applied to the application environment shown in fig. 1. Wherein the terminal 102 communicates with the server 104 via a network. The terminal 102 may obtain the corresponding video stream data from the server 104 by sending a request to the server 104. Specifically, the terminal 102 sends a data acquisition request to the server 104, so that the server 104 queries corresponding video stream data according to the received data acquisition request, and sequentially transmits the corresponding video stream data to the terminal 102, where the video stream data includes multiple frames of images. The terminal 102 reads the multi-frame image, detects a corresponding target in the multi-frame image, and generates a sample image set of a plurality of categories using the target. The terminal 102 trains the sample image sets of the multiple categories by using the recognition model to obtain recognition probabilities corresponding to the categories. The terminal 102 adjusts the samples in the corresponding sample image sets according to the recognition probability, the terminal 102 optimizes the recognition models by using the adjusted sample image sets, and the optimized recognition models are trained through the adjusted sample image sets. The terminal 102 may be, but is not limited to, various personal computers, notebook computers, smart phones, and tablet computers, and the server 104 may be implemented by an independent server or a server cluster formed by a plurality of servers.
In an embodiment, as shown in fig. 2, a training method for target recognition is provided, which is described by taking the method as an example applied to the terminal in fig. 1, and includes the following steps:
The server stores a large amount of video stream data, and the terminal can acquire the video stream data from the server and perform video decoding on the acquired video stream data to obtain a multi-frame image with a uniform picture format. The server can also directly transmit the stored video stream data to the terminal in sequence. Specifically, the terminal sends a data acquisition request to the server according to different requirements of the user, so that the server queries corresponding video stream data according to the received data acquisition request, and transmits the corresponding video stream data to the terminal in sequence. The video stream data comprises a plurality of frames of images which are arranged in sequence, and the transmission of the video stream data refers to that the plurality of frames of images are transmitted through the video stream in sequence.
After the terminal receives video stream data transmitted by the server, the terminal reads a plurality of frames of images in the video stream data, detects corresponding targets in the plurality of frames of images, and generates a plurality of categories of sample image sets according to the detected targets. For example, the terminal reads multi-frame images in video stream data, detects that corresponding target objects are felines in the multi-frame images, and generates corresponding feline sample image sets of multiple categories according to different types of the felines.
And the terminal trains the sample image sets of the multiple categories by using the recognition model according to the generated sample image sets of the multiple categories to obtain recognition probabilities corresponding to the categories. Specifically, the terminal may preset a standard sample number value according to an experimental test result, the terminal invokes the calculation module to calculate the number of images of the sample image set of each category to obtain a sample number corresponding to the sample image set of each category, and the terminal calculates a ratio between the sample number of the sample image set of each category and the preset standard sample number value to generate sample proportions of the sample image sets of the corresponding multiple categories. The terminal can obtain the sample proportion of the sample image sets of the multiple categories, the terminal scans the sample proportion of the sample image sets of the multiple categories in a traversing mode by using the configuration tool, the terminal compares the scanned sample proportion with the preset threshold value, and the terminal can divide the sample image sets of the multiple categories into sample image sets of a small number of categories and sample image sets of a large number of categories according to different preset threshold value ranges. Further, the terminal trains the sample image set of the minority class by using the recognition model to obtain the recognition probability corresponding to the minority class. And the terminal trains the sample image sets of the plurality of categories by using the identification model to obtain the identification probabilities corresponding to the plurality of categories. For example, when the terminal detects that the sample proportion is not greater than 0.5, the terminal marks the sample image set of the category corresponding to the sample proportion as the sample image set of the few categories, which indicates that the number of samples of the sample image set of the category is seriously missing, and an overfitting phenomenon may occur in a single enhancement processing mode. When the terminal detects that the sample proportion is not less than 1.5, the terminal marks the sample image set of the category corresponding to the sample proportion as the sample image sets of the majority of categories, which indicates that the number of samples of the sample image set is too large, and a single deleted sample can reduce the characteristic information and reduce the recognition rate, so that an under-fitting phenomenon can occur.
And step 208, adjusting the samples in the corresponding sample image set according to the recognition probability.
And step 210, optimizing the recognition model by using the adjusted sample image set, and training the optimized recognition model by using the adjusted sample image set.
And the terminal adjusts the samples in the corresponding sample image set according to the recognition probability. Specifically, the terminal trains a sample image set of a few categories by using the recognition model, records the recognition result of the sample image set after each training, and obtains the recognition probability corresponding to the few categories. And the terminal adjusts the sample image set of the minority class according to the identification probability corresponding to the minority class to obtain the adjusted sample image set. And the terminal trains the sample image sets of the majority of categories by using the identification model, records the identification result of the sample image set after each training, and obtains the identification probability corresponding to the majority of categories. And the terminal adjusts the sample image sets of the majority categories according to the identification probabilities corresponding to the majority categories to obtain the adjusted sample image sets. The identification result of the sample image set after each training can include target coordinate information corresponding to the sample image set of each category. And the terminal adjusts the samples in the corresponding sample image set according to the identification probability, wherein the adjustment comprises enhancing or cutting the sample image set to obtain the adjusted sample image set. And the terminal optimizes the recognition model by using the adjusted sample image set, and trains the optimized recognition model through the adjusted sample image set.
In this embodiment, by acquiring video stream data, the video stream data includes multiple frames of images. Reading the image, detecting a corresponding target in the image, and generating a sample image set of a plurality of categories by using the target. And training the sample image sets of the multiple categories by using the recognition model to obtain recognition probabilities corresponding to the categories. And adjusting the samples in the corresponding sample image set according to the recognition probability. And optimizing the recognition model by using the adjusted sample image set, and training the optimized recognition model by using the adjusted sample image set. Compared with the traditional target recognition training mode, the method has the advantages that the number relation of samples among different types of sample image sets is analyzed, a small number of types of samples is enhanced, a large number of types of samples are weakened, efficient and organic fusion application is achieved, the effect of balancing the number of samples among the types in the sample library is achieved, meanwhile, the learning training quality of the neural network is not affected, namely, the problem of poor model learning effect caused by unbalanced samples is solved by adopting a dynamic processing scheme according to the number relation of the samples among the different types of sample image sets in the sample library, therefore, the trained recognition model can carry out accurate target recognition, and the problems of under-fitting and over-fitting of the training model are avoided.
In one embodiment, the step of training the sample image set of multiple classes using the recognition model, as shown in fig. 3, comprises:
In step 304, the minimum number of samples is selected as the standard unit amount.
And step 306, equally dividing the number of samples of the sample image set of each category according to the standard unit amount to obtain a plurality of corresponding standard unit amount sample image sets.
The terminal reads the multi-frame images, detects corresponding targets in the multi-frame images, and generates a sample image set of multiple categories by using the targets. Further, the terminal trains the sample image sets of multiple categories by using the recognition model. Specifically, the terminal obtains the number of samples corresponding to the sample image set of each category, and selects the minimum number of samples as the standard unit number. And the terminal equally divides the number of samples of the sample image set of each category to be trained according to the standard unit amount to obtain a plurality of corresponding standard unit amount sample image sets. For example, the terminal obtains the number of samples corresponding to the sample image set of each category, and selects the minimum number Ns of samples as the standard unit number. If the number of samples of the sample image sets of one category is Nm, the terminal calculates the multiple relation between the number of samples Nm and the standard unit amount Ns, that is, Nm/Ns is m, where m is rounded up, and then the terminal divides the sample image sets corresponding to the number of samples Nm into m standard unit amount sample image sets. Further, the terminal trains the standard unit quantity sample image set corresponding to the sample image set of each category by using a K-fold cross validation method, records the identification result of the sample image set after each training, and obtains the identification probability corresponding to the category. For example, the sample image library has 3 sample image sets in a first category, 10 sample images in a first sample image set, 21 sample images in a second sample image set, and 38 sample images in a third sample image set. The terminal acquires the number of samples corresponding to the sample image set of the 3 categories, i.e., N1 is 10, N2 is 21, and N3 is 38, and selects the sample amount N1 of the sample image set of which the minimum sample amount is the first category as 10, i.e., Ns1 is 10 as a standard unit amount. Further, the terminal equally divides the second category of sample image sets according to the standard unit amount Ns, namely, the second category of sample image sets is equally divided into 2 standard unit amount sample image sets Ns (10) and Ns (11), and then the terminal takes the Ns (10) standard unit amount sample image set as a training sample, and takes the remaining Ns (11) standard unit amount sample image set as a test sample for training. And the terminal equally divides the sample image set of the third category according to the standard unit amount Ns, namely the sample image set of the third category is equally divided into 4 standard unit amount sample image sets Ns (10), Ns (10) and Ns (8), and the terminal takes the Ns (10) standard unit amount sample image set as a training sample and takes the rest 3 Ns (10), Ns (10) and Ns (8) standard unit amount sample image sets as test samples for training. Therefore, by using the K-fold cross validation method, the sample image set of each category is equally divided into a plurality of standard unit quantity sample image sets, the chance that each sample is only once classified into a training set or a test set in each iteration process is realized, the problem of poor model learning effect caused by unbalanced samples is solved, and the problems of under-fitting and over-fitting of the training model are avoided.
In one embodiment, when the terminal trains the standard unit amount image sets corresponding to the sample image sets of each category, if there are a plurality of standard unit amount image sets corresponding to the sample image sets of one category, a K-fold cross validation method may be used, and the terminal may use one of the standard unit amount image sets as a training sample and use the remaining standard unit amount image sets as test samples. For example, the terminal obtains the sample numbers N1, N2, and N3 corresponding to the sample image sets of each category, and selects the sample size of the sample image set of which the minimum sample number is the third category, that is, Ns — N3 is a standard unit size. The standard unit amount image set corresponding to the sample image set of the 1 st category is (Ns11, Ns12, Ns13), the standard unit amount image set corresponding to the sample image set of the 2 nd category is (Ns21, Ns22), the standard unit amount image set corresponding to the sample image set of the 3 rd category is (Ns31), because Ns31 is the minimum sample number, namely the standard unit amount, in each iteration process, the terminal does not repeatedly extract 20% of images from the standard unit amount image set corresponding to the sample image set of the 3 rd category as a test set, and the rest 80% of images are used as a training set, namely the test set corresponding to the sample image set of the 3 rd category is 0.2 (Ns31), and the training set is 0.8 (Ns 31). And the terminal sequentially performs the following combination training according to the permutation and combination principle. The terminal trains the test set { Ns12, Ns13, Ns22,0.2 x (Ns31) } according to the first training set { Ns11, Ns21, 0.8 x (Ns31) }. The terminal trains the test set { Ns12, Ns13, Ns21,0.2 x (Ns31) } according to the second training set { Ns11, Ns22,0.8 x (Ns31) }. The terminal trains the test set { Ns11, Ns13, Ns22,0.8 × (Ns31) } according to the third training set { Ns12, Ns21, 0.8 × (Ns31) }. The terminal trains the test set { Ns11, Ns13, Ns21,0.2 x (Ns31) } according to the fourth training set { Ns12, Ns22,0.8 x (Ns31) }. The terminal trains the test set { Ns11, Ns12, Ns22,0.2 x (Ns31) } according to the fifth training set { Ns13, Ns21, 0.8 x (Ns31) }. And the terminal trains the test set of { Ns11, Ns12, Ns21 and 0.2 (Ns31) }accordingto a sixth training set of { Ns13, Ns22 and 0.8 (Ns31) }, and records the identification result of the sample image set after each training, namely the identification probability corresponding to the test set after each training. Therefore, the recognition model can train different types of sample image sets according to the standard unit amount of the sample image sets, and the dynamic sample balance problem of multiple samples can be effectively processed.
In one embodiment, the step of performing an adjustment process on the samples in the corresponding sample image set according to the recognition probability, as shown in fig. 4, includes:
And step 404, when the sample proportion reaches a first threshold value, marking the sample image set of the category corresponding to the sample proportion as the sample image set of the majority category.
And 408, performing similarity matching on the target coordinate accuracy and the preset target coordinate accuracy to obtain the recognition probability corresponding to the majority of categories.
And step 410, when the identification probability is greater than the preset identification probability, performing cutting processing on the sample image sets of the majority of categories corresponding to the identification probability until the number of samples corresponding to the sample image sets of the majority of categories reaches a data balance reasonable range.
The terminal trains the sample image sets of multiple categories by using the recognition model to obtain recognition probabilities corresponding to the categories, and the terminal adjusts the samples in the corresponding sample image sets according to the recognition probabilities. Specifically, before the terminal trains the sample image sets of multiple categories by using the recognition model, the terminal may obtain the sample proportions of the sample image sets of multiple categories, and the terminal may scan the sample proportions of the sample image sets of multiple categories by using a configuration tool in a traversal manner, and when the terminal detects that the sample proportions reach a first threshold value, the sample image sets of the categories corresponding to the sample proportions are labeled as the sample image sets of the majority categories. When the terminal loads the recognition model to train the sample image sets of the multiple categories, the terminal trains the standard unit amount sample image sets corresponding to the sample image sets of the multiple categories in sequence according to the permutation and combination principle, and the recognition result of the test set after each training is recorded. Further, the terminal obtains the accuracy of the target coordinate corresponding to each test set according to the recognition result of each test set after training. The terminal carries out image labeling on multi-frame images corresponding to the target in advance to obtain preset target coordinate accuracy, and similarity matching is carried out on the target coordinate accuracy corresponding to each obtained test set and the preset target coordinate accuracy by the terminal to obtain the recognition probability of a plurality of test sets corresponding to a plurality of categories. And when the terminal detects that the identification probability of one test set is greater than the preset identification probability, the terminal performs cutting processing on the sample image sets of the majority categories corresponding to the test set until the number of samples corresponding to the sample image sets of the majority categories reaches a reasonable sample data balance range. Therefore, the sample images which are easy to identify in the samples of the plurality of types are found by utilizing the deep learning algorithm and are subjected to data reduction, the image data characteristics of the samples of the plurality of types are not lost too much, meanwhile, the samples of the plurality of types can be cut appropriately, finally, the unbalanced sample set is dynamically adjusted to be relatively balanced, and the problem of poor model learning effect caused by unbalanced samples is effectively solved.
In one embodiment, the step of performing an adjustment process on the samples in the corresponding sample image set according to the recognition probability, as shown in fig. 5, includes:
And step 504, when the sample proportion reaches a second threshold value, marking the sample image set of the class corresponding to the sample proportion as the sample image set of the few classes.
And step 508, performing similarity matching on the target coordinate accuracy and the preset target coordinate accuracy to obtain the recognition probability corresponding to a few categories.
And 510, when the identification probability is smaller than the preset identification probability, performing enhancement processing on the sample image sets of the few categories corresponding to the identification probability until the number of samples corresponding to the sample image sets of the few categories reaches a data balance reasonable range.
The terminal trains the sample image sets of multiple categories by using the recognition model to obtain recognition probabilities corresponding to the categories, and the terminal adjusts the samples in the corresponding sample image sets according to the recognition probabilities. Specifically, before the terminal trains the sample image sets of multiple categories by using the recognition model, the terminal may obtain the sample proportions of the sample image sets of multiple categories, and the terminal may scan the sample proportions of the sample image sets of multiple categories by using a configuration tool in a traversal manner, and when the terminal detects that the sample proportions reach a second threshold, the sample image sets of the categories corresponding to the sample proportions are labeled as the sample image sets of a few categories. And when the terminal loads the identification model to train the sample image sets of the minority categories, the terminal trains the standard unit amount sample image sets corresponding to the sample image sets of the minority categories in sequence according to the permutation and combination principle, and the identification result of the test set after each training is recorded. Further, the terminal obtains the accuracy of the target coordinate corresponding to each test set according to the recognition result of each test set after training. The terminal carries out image labeling on multi-frame images corresponding to the target in advance to obtain preset target coordinate accuracy, and similarity matching is carried out on the target coordinate accuracy corresponding to each obtained test set and the preset target coordinate accuracy by the terminal to obtain the recognition probability of a plurality of test sets corresponding to a few categories. When the terminal detects that the identification probability of one test set is smaller than the preset identification probability, the terminal performs enhancement processing on the sample image sets of the few categories corresponding to the test set until the number of samples corresponding to the sample image sets of the few categories reaches a reasonable sample data balance range. Therefore, the sample images which are difficult to identify in a few types of samples are found out by utilizing a deep learning algorithm, data enhancement processing is carried out on the sample images which are difficult to identify, finally, the unbalanced sample set is dynamically adjusted to be relatively balanced, and the problem of poor model learning effect caused by unbalanced samples is effectively solved.
In one embodiment, the step of performing enhancement processing on the sample image set of the minority class corresponding to the identification probability includes:
and randomly extracting the sample images in the sample image set of the few categories to obtain randomly extracted original sample images.
And randomly adjusting the brightness and the contrast of the original sample image to obtain a newly-added sample image.
And carrying out shielding processing on the original sample image to obtain a newly added sample image.
When the terminal detects that the recognition probability of one test set is smaller than the preset recognition probability, the terminal performs enhancement processing on the sample image sets of the few categories corresponding to the test set until the number of samples corresponding to the sample image sets of the few categories reaches sample balance. Specifically, the terminal randomly extracts images in the sample image set of a few categories corresponding to the test set to obtain randomly extracted original sample images. The terminal may preset a maximum contrast threshold m1 and a maximum bias threshold m2, and randomly generate a random number n within 10, convert it to a percentage value p n/100, according to the following adjustment equation:
g(x)＝af(x)+b (1)
wherein: (x) is the source image; g (x) is an output image; a is a gain value used for setting image contrast; b is an offset value used for adjusting the image brightness.
The terminal calculates according to the following formula:
a＝m1*p/100 (2)
b＝m2*p (3)
and (3) substituting the calculated a and b into the formula (1) by the terminal to obtain a newly added sample image with randomly adjusted brightness and contrast. The randomness of the image brightness and the contrast is enhanced through the random processing of the image brightness and the contrast, so that the richness of the sample is improved, and the robustness of the neural network is improved.
The terminal can also carry out random shielding processing on the original sample image to obtain a newly added sample image after random shielding. For example, the terminal may uniformly divide the original sample image into 16 image units according to 4 × 4, and randomly occlude 1 or 2 image units of the 16 image units, to obtain a randomly occluded new sample image. Therefore, the few category samples lacking the number of samples are subjected to enhancement processing, and the enhancement processing is performed from the aspects of color, shielding and the like by adopting the sampling method, so that the diversity and the robustness of the samples are improved, and the training recognition model can better learn the image characteristics of the samples.
In one embodiment, the step of optimizing the recognition model using the adjusted sample image set comprises:
and optimizing the recognition model by adjusting a loss function in the recognition model and adding the adjusted sample quantity proportional weight of the sample image set to the recognition model.
And the terminal adjusts the samples in the corresponding sample image set according to the recognition probability, optimizes the recognition model by using the adjusted sample image set, and trains the optimized recognition model through the adjusted sample image set. Specifically, the terminal may calculate a loss function value of each adjusted sample image set, modify the loss function value in the identification model according to the calculated loss function value corresponding to each adjusted sample image set, add a sample number proportional weight of each adjusted sample image set to the identification model, and optimize the identification model. The terminal trains the optimized recognition model through the adjusted sample image sets, so that when the target recognition of the unbalanced sample image set is carried out, the classification accuracy of the sample image sets of a few categories is improved by adding the sample quantity proportion weight of each adjusted sample image set to the recognition model, and the trained recognition model can carry out accurate target recognition.
It should be understood that although the various steps in the flow charts of fig. 1-5 are shown in order as indicated by the arrows, the steps are not necessarily performed in order as indicated by the arrows. The steps are not performed in the exact order shown and described, and may be performed in other orders, unless explicitly stated otherwise. Moreover, at least some of the steps in fig. 1-5 may include multiple sub-steps or multiple stages that are not necessarily performed at the same time, but may be performed at different times, and the order of performance of the sub-steps or stages is not necessarily sequential, but may be performed in turn or alternating with other steps or at least some of the sub-steps or stages of other steps.
In one embodiment, as shown in fig. 6, there is provided a training apparatus for object recognition, including: an obtaining module 602, a generating module 604, a training module 606, an adjusting module 608, and an optimizing module 610, wherein:
the obtaining module 602 is configured to obtain video stream data, where the video stream data includes multiple frames of images.
The generating module 604 is configured to read the image, detect a corresponding object in the image, and generate a sample image set of multiple categories using the object.
The training module 606 is configured to train the sample image sets of multiple categories by using the recognition model, so as to obtain recognition probabilities corresponding to the categories.
And an adjusting module 608, configured to perform an adjusting process on the samples in the corresponding sample image set according to the recognition probability.
And the optimizing module 610 is configured to optimize the recognition model by using the adjusted sample image set, and train the optimized recognition model through the adjusted sample image set.
In one embodiment, the apparatus further comprises: a selection module and an equal division module.
The obtaining module 602 is further configured to obtain a number of samples corresponding to the sample image set of each category. The selection module is used for selecting the minimum sample number as the standard unit quantity. The dividing module is used for dividing the number of samples of the sample image set of each category equally according to the standard unit amount to obtain a plurality of corresponding standard unit amount sample image sets. The training module 606 is further configured to train a standard unit amount sample image set corresponding to the sample image set of each category.
In one embodiment, the apparatus further comprises: the device comprises a marking module, a matching module and a cutting processing module.
The obtaining module 602 is further configured to obtain a sample scale corresponding to the sample image set. The marking module is used for marking the sample image set of the category corresponding to the sample proportion as the sample image set of the majority category when the sample proportion reaches a first threshold value. The obtaining module 602 is further configured to obtain target coordinate accuracies corresponding to the plurality of categories of sample image sets. The matching module is used for carrying out similarity matching on the target coordinate accuracy and the preset target coordinate accuracy to obtain the recognition probability corresponding to most categories. And the cutting processing module is used for cutting the sample image sets of the majority categories corresponding to the recognition probability when the recognition probability is greater than the preset recognition probability until the number of samples corresponding to the sample image sets of the majority categories reaches a data balance reasonable range.
In one embodiment, the apparatus further comprises: and an enhancement processing module.
The obtaining module 602 is further configured to obtain a sample scale corresponding to the sample image set. The marking module is further used for marking the sample image set with the sample scale corresponding to the category as the sample image set with the few categories when the sample scale reaches a second threshold value. The obtaining module 602 is further configured to obtain target coordinate accuracy corresponding to the sample image set of the few categories. The matching module is also used for carrying out similarity matching on the target coordinate accuracy and the preset target coordinate accuracy to obtain the recognition probability corresponding to a few categories. And the enhancement processing module is used for enhancing the sample image sets of the few categories corresponding to the recognition probability when the recognition probability is smaller than the preset recognition probability until the number of samples corresponding to the sample image sets of the few categories reaches a data balance reasonable range.
In one embodiment, the apparatus further comprises: the device comprises an extraction module and a shielding processing module.
The extraction module is used for randomly extracting the sample images in the sample image set of the few categories to obtain randomly extracted original sample images. The adjusting module 608 is further configured to randomly adjust the brightness and the contrast of the original sample image to obtain a new sample image. And the shielding processing module is used for shielding the original sample image to obtain a newly added sample image.
In one embodiment, the adjustment module 608 is further configured to optimize the recognition model by adjusting a loss function in the recognition model and adding a sample number proportional weight of the adjusted sample image set to the recognition model.
For the specific definition of the training device for target recognition, reference may be made to the above definition of the training method for target recognition, which is not described herein again. The modules in the training device for target recognition can be wholly or partially realized by software, hardware and a combination thereof. The modules can be embedded in a hardware form or independent from a processor in the computer device, and can also be stored in a memory in the computer device in a software form, so that the processor can call and execute operations corresponding to the modules.
In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as shown in fig. 7. The computer device includes a processor, a memory, a network interface, a display screen, and an input device connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of an operating system and computer programs in the non-volatile storage medium. The network interface of the computer device is used for communicating with an external terminal through a network connection. The computer program is executed by a processor to implement a training method for object recognition. The display screen of the computer equipment can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer equipment can be a touch layer covered on the display screen, a key, a track ball or a touch pad arranged on the shell of the computer equipment, an external keyboard, a touch pad or a mouse and the like.
Those skilled in the art will appreciate that the architecture shown in fig. 7 is merely a block diagram of some of the structures associated with the disclosed aspects and is not intended to limit the computing devices to which the disclosed aspects apply, as particular computing devices may include more or less components than those shown, or may combine certain components, or have a different arrangement of components.
In one embodiment, a computer device is provided, comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the steps of the above-described method embodiments being implemented when the computer program is executed by the processor.
It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by hardware instructions of a computer program, which can be stored in a non-volatile computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. Any reference to memory, storage, database, or other medium used in the embodiments provided herein may include non-volatile and/or volatile memory, among others. Non-volatile memory can include read-only memory (ROM), Programmable ROM (PROM), Electrically Programmable ROM (EPROM), Electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDRSDRAM), Enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus Direct RAM (RDRAM), direct bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
The technical features of the above embodiments can be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the above embodiments are not described, but should be considered as the scope of the present specification as long as there is no contradiction between the combinations of the technical features.
The above-mentioned embodiments only express several embodiments of the present application, and the description thereof is more specific and detailed, but not construed as limiting the scope of the invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the concept of the present application, which falls within the scope of protection of the present application. Therefore, the protection scope of the present patent shall be subject to the appended claims.
Claims (10)
1. A training method of object recognition, the method comprising:
acquiring video stream data, wherein the video stream data comprises a plurality of frames of images;
reading the image, detecting a corresponding target in the image, and generating a sample image set of a plurality of categories by using the target;
training the sample image sets of multiple categories by using an identification model to obtain identification probabilities corresponding to the categories;
adjusting the samples in the corresponding sample image set according to the identification probability;
and optimizing the recognition model by using the adjusted sample image set, and training the optimized recognition model by using the adjusted sample image set.
2. The method of claim 1, wherein training the sample image set for a plurality of classes using a recognition model comprises:
acquiring the number of samples corresponding to the sample image set of each category;
selecting the minimum sample number as a standard unit amount;
equally dividing the number of samples of the sample image set of each category according to the standard unit amount to obtain a plurality of corresponding standard unit amount sample image sets;
and training the standard unit quantity sample image set corresponding to the sample image set of each category.
3. The method of claim 1, wherein the adjusting the samples in the corresponding sample image set according to the recognition probability comprises:
acquiring a sample proportion corresponding to the sample image set;
when the sample proportion reaches a first threshold value, marking the sample image set of the category corresponding to the sample proportion as a sample image set of a majority category;
acquiring target coordinate accuracy corresponding to the sample image sets of the multiple categories;
similarity matching is carried out on the target coordinate accuracy and preset target coordinate accuracy to obtain recognition probabilities corresponding to most categories;
and when the identification probability is greater than the preset identification probability, performing clipping processing on the sample image sets of the majority categories corresponding to the identification probability until the number of samples corresponding to the sample image sets of the majority categories reaches a data balance reasonable range.
4. The method of claim 1, wherein the adjusting the samples in the corresponding sample image set according to the recognition probability comprises:
acquiring a sample proportion corresponding to the sample image set;
when the sample proportion reaches a second threshold value, marking the sample image set of the class corresponding to the sample proportion as the sample image set of the minority class;
acquiring target coordinate accuracy corresponding to the sample image sets of the minority categories;
similarity matching is carried out on the target coordinate accuracy and preset target coordinate accuracy, and recognition probabilities corresponding to a few categories are obtained;
and when the identification probability is smaller than the preset identification probability, performing enhancement processing on the sample image sets of the minority categories corresponding to the identification probability until the number of samples corresponding to the sample image sets of the minority categories reaches a data balance reasonable range.
5. The method according to claim 4, wherein the enhancing the sample image set of the minority class corresponding to the recognition probability comprises:
randomly extracting the sample images in the sample image set of the minority class to obtain original randomly extracted sample images;
randomly adjusting the brightness and the contrast of the original sample image to obtain a newly added sample image;
and carrying out shielding processing on the original sample image to obtain a newly added sample image.
6. The method of claim 1, wherein optimizing the recognition model using the adjusted sample image set comprises:
and optimizing the recognition model by adjusting a loss function in the recognition model and adding the adjusted sample quantity proportional weight of the sample image set to the recognition model.
7. An apparatus for training object recognition, the apparatus comprising:
the acquisition module is used for acquiring video stream data, and the video stream data comprises a plurality of frames of images;
the generating module is used for reading the image, detecting a corresponding target in the image, and generating a sample image set of a plurality of categories by using the target;
the training module is used for training the sample image sets of multiple categories by using a recognition model to obtain recognition probabilities corresponding to the categories;
the adjusting module is used for adjusting the samples in the corresponding sample image set according to the identification probability;
and the optimization module is used for optimizing the recognition model by utilizing the adjusted sample image set and training the optimized recognition model through the adjusted sample image set.
8. Training device for object recognition according to claim 7, characterized in that it further comprises:
the obtaining module is further used for obtaining the number of samples corresponding to the sample image set of each category;
the selection module is used for selecting the minimum sample number as the standard unit number;
the dividing module is used for dividing the number of samples of the sample image set of each category equally according to the standard unit amount to obtain a plurality of corresponding standard unit amount sample image sets;
the training module is further used for training the standard unit quantity sample image set corresponding to the sample image set of each category.
9. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the steps of the method of any of claims 1 to 6 are implemented when the computer program is executed by the processor.
10. 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 method of any one of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911300954.3A CN111079841A (en) | 2019-12-17 | 2019-12-17 | Training method and device for target recognition, computer equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911300954.3A CN111079841A (en) | 2019-12-17 | 2019-12-17 | Training method and device for target recognition, computer equipment and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111079841A true CN111079841A (en) | 2020-04-28 |
Family
ID=70315086
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911300954.3A Pending CN111079841A (en) | 2019-12-17 | 2019-12-17 | Training method and device for target recognition, computer equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111079841A (en) |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111625636A (en) * | 2020-05-28 | 2020-09-04 | 深圳追一科技有限公司 | Man-machine conversation refusal identification method, device, equipment and medium |
| CN112149745A (en) * | 2020-09-27 | 2020-12-29 | 上海高德威智能交通系统有限公司 | Method, device, device and storage medium for determining difficult case samples |
| CN112163637A (en) * | 2020-10-19 | 2021-01-01 | 平安国际智慧城市科技股份有限公司 | Image classification model training method and device based on unbalanced data |
| CN112183321A (en) * | 2020-09-27 | 2021-01-05 | 深圳奇迹智慧网络有限公司 | Method and device for optimizing machine learning model, computer equipment and storage medium |
| CN112348051A (en) * | 2020-09-30 | 2021-02-09 | 北京信工博特智能科技有限公司 | Method and system for reducing sample imbalance influence in target detection |
| CN112560900A (en) * | 2020-09-08 | 2021-03-26 | 同济大学 | Multi-disease classifier design method for sample imbalance |
| CN112712136A (en) * | 2021-01-15 | 2021-04-27 | 清华大学 | Method and device for intelligently manufacturing sand surface image-mud content relation data set |
| CN112836768A (en) * | 2021-03-08 | 2021-05-25 | 北京电子工程总体研究所 | Data balancing method and system, computer equipment and medium |
| CN113361487A (en) * | 2021-07-09 | 2021-09-07 | 无锡时代天使医疗器械科技有限公司 | Foreign matter detection method, device, equipment and computer readable storage medium |
| CN113505780A (en) * | 2021-08-05 | 2021-10-15 | 青岛新奥清洁能源有限公司 | Two-dimensional code-based intelligent detection maintenance method and equipment |
| CN114330364A (en) * | 2021-12-27 | 2022-04-12 | 北京百度网讯科技有限公司 | Model training method, intention recognition device and electronic equipment |
| CN114494961A (en) * | 2022-01-24 | 2022-05-13 | 广州富港生活智能科技有限公司 | Sample data processing method and device applied to model training |
| WO2022121289A1 (en) * | 2020-12-11 | 2022-06-16 | Huawei Cloud Computing Technologies Co., Ltd. | Methods and systems for mining minority-class data samples for training neural network |
| CN114677541A (en) * | 2022-03-23 | 2022-06-28 | 成都智元汇信息技术股份有限公司 | Method and system for extracting adhesion sample set based on target |
| CN115660596A (en) * | 2022-11-03 | 2023-01-31 | 创启科技(广州)有限公司 | Data interaction method of mobile terminal and mobile terminal |
| CN119963913A (en) * | 2025-01-13 | 2025-05-09 | 北京市生态环境保护科学研究院 | A method, device and medium for summarizing biodiversity image data |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107784312A (en) * | 2016-08-24 | 2018-03-09 | 腾讯征信有限公司 | Machine learning model training method and device |
| US10402691B1 (en) * | 2018-10-04 | 2019-09-03 | Capital One Services, Llc | Adjusting training set combination based on classification accuracy |
| CN110443280A (en) * | 2019-07-05 | 2019-11-12 | 北京达佳互联信息技术有限公司 | Training method, device and the storage medium of image detection model |
-
2019
- 2019-12-17 CN CN201911300954.3A patent/CN111079841A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107784312A (en) * | 2016-08-24 | 2018-03-09 | 腾讯征信有限公司 | Machine learning model training method and device |
| US10402691B1 (en) * | 2018-10-04 | 2019-09-03 | Capital One Services, Llc | Adjusting training set combination based on classification accuracy |
| CN110443280A (en) * | 2019-07-05 | 2019-11-12 | 北京达佳互联信息技术有限公司 | Training method, device and the storage medium of image detection model |
Cited By (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111625636A (en) * | 2020-05-28 | 2020-09-04 | 深圳追一科技有限公司 | Man-machine conversation refusal identification method, device, equipment and medium |
| CN112560900A (en) * | 2020-09-08 | 2021-03-26 | 同济大学 | Multi-disease classifier design method for sample imbalance |
| CN112560900B (en) * | 2020-09-08 | 2023-01-20 | 同济大学 | Multi-disease classifier design method for sample imbalance |
| CN112149745A (en) * | 2020-09-27 | 2020-12-29 | 上海高德威智能交通系统有限公司 | Method, device, device and storage medium for determining difficult case samples |
| CN112183321A (en) * | 2020-09-27 | 2021-01-05 | 深圳奇迹智慧网络有限公司 | Method and device for optimizing machine learning model, computer equipment and storage medium |
| CN112149745B (en) * | 2020-09-27 | 2022-10-21 | 上海高德威智能交通系统有限公司 | Method, device, device and storage medium for determining difficult case samples |
| CN112348051A (en) * | 2020-09-30 | 2021-02-09 | 北京信工博特智能科技有限公司 | Method and system for reducing sample imbalance influence in target detection |
| CN112163637B (en) * | 2020-10-19 | 2024-04-19 | 深圳赛安特技术服务有限公司 | Image classification model training method and device based on unbalanced data |
| CN112163637A (en) * | 2020-10-19 | 2021-01-01 | 平安国际智慧城市科技股份有限公司 | Image classification model training method and device based on unbalanced data |
| US11816183B2 (en) | 2020-12-11 | 2023-11-14 | Huawei Cloud Computing Technologies Co., Ltd. | Methods and systems for mining minority-class data samples for training a neural network |
| WO2022121289A1 (en) * | 2020-12-11 | 2022-06-16 | Huawei Cloud Computing Technologies Co., Ltd. | Methods and systems for mining minority-class data samples for training neural network |
| CN112712136A (en) * | 2021-01-15 | 2021-04-27 | 清华大学 | Method and device for intelligently manufacturing sand surface image-mud content relation data set |
| CN112836768A (en) * | 2021-03-08 | 2021-05-25 | 北京电子工程总体研究所 | Data balancing method and system, computer equipment and medium |
| CN113361487A (en) * | 2021-07-09 | 2021-09-07 | 无锡时代天使医疗器械科技有限公司 | Foreign matter detection method, device, equipment and computer readable storage medium |
| CN113505780A (en) * | 2021-08-05 | 2021-10-15 | 青岛新奥清洁能源有限公司 | Two-dimensional code-based intelligent detection maintenance method and equipment |
| CN114330364A (en) * | 2021-12-27 | 2022-04-12 | 北京百度网讯科技有限公司 | Model training method, intention recognition device and electronic equipment |
| CN114330364B (en) * | 2021-12-27 | 2022-11-11 | 北京百度网讯科技有限公司 | Model training method, intention recognition device and electronic equipment |
| CN114494961A (en) * | 2022-01-24 | 2022-05-13 | 广州富港生活智能科技有限公司 | Sample data processing method and device applied to model training |
| CN114677541A (en) * | 2022-03-23 | 2022-06-28 | 成都智元汇信息技术股份有限公司 | Method and system for extracting adhesion sample set based on target |
| CN115660596A (en) * | 2022-11-03 | 2023-01-31 | 创启科技(广州)有限公司 | Data interaction method of mobile terminal and mobile terminal |
| CN115660596B (en) * | 2022-11-03 | 2023-07-04 | 创启科技(广州)有限公司 | Data interaction method of mobile terminal and mobile terminal |
| CN119963913A (en) * | 2025-01-13 | 2025-05-09 | 北京市生态环境保护科学研究院 | A method, device and medium for summarizing biodiversity image data |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111079841A (en) | Training method and device for target recognition, computer equipment and storage medium | |
| CN110889457B (en) | Sample image classification training method and device, computer equipment and storage medium | |
| US11710293B2 (en) | Target detection method and apparatus, computer-readable storage medium, and computer device | |
| CN110930417B (en) | Training method and device for image segmentation model, and image segmentation method and device | |
| CN108377240B (en) | Abnormal interface detection method and device, computer equipment and storage medium | |
| EP2806374B1 (en) | Method and system for automatic selection of one or more image processing algorithm | |
| CN109285105B (en) | Watermark detection method, watermark detection device, computer equipment and storage medium | |
| CN110019790B (en) | Text recognition, text monitoring, data object recognition and data processing method | |
| WO2023050650A1 (en) | Animation video generation method and apparatus, and device and storage medium | |
| WO2022126917A1 (en) | Deep learning-based face image evaluation method and apparatus, device, and medium | |
| CN112183296B (en) | Simulated bill image generation and bill image recognition method and device | |
| CN111832581A (en) | Lung feature recognition method and device, computer equipment and storage medium | |
| US20230177251A1 (en) | Method, device, and system for analyzing unstructured document | |
| CN111783812A (en) | Method and device for identifying forbidden images and computer readable storage medium | |
| CN113961768B (en) | Sensitive word detection method and device, computer equipment and storage medium | |
| WO2023173546A1 (en) | Method and apparatus for training text recognition model, and computer device and storage medium | |
| CN112163110A (en) | Image classification method and device, electronic equipment and computer-readable storage medium | |
| CN113869099A (en) | Image processing method and device, electronic equipment and storage medium | |
| CN112464660A (en) | Text classification model construction method and text data processing method | |
| CN112396047A (en) | Training sample generation method and device, computer equipment and storage medium | |
| CN113283388B (en) | Training method, device, equipment and storage medium of living body face detection model | |
| CN112348060A (en) | Classification vector generation method and device, computer equipment and storage medium | |
| CN115359490A (en) | Text direction identification method, device, equipment and storage medium | |
| CN114782994A (en) | Gesture recognition method and device, storage medium and electronic equipment | |
| CN114116512A (en) | Test methods, apparatus, electronic equipment and storage media |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200428 |
|
| RJ01 | Rejection of invention patent application after publication |