CN111242211B - Underground intrusion signal identification method and system in field cultural relic protection system - Google Patents
Underground intrusion signal identification method and system in field cultural relic protection system Download PDFInfo
- Publication number
- CN111242211B CN111242211B CN202010029383.0A CN202010029383A CN111242211B CN 111242211 B CN111242211 B CN 111242211B CN 202010029383 A CN202010029383 A CN 202010029383A CN 111242211 B CN111242211 B CN 111242211B
- Authority
- CN
- China
- Prior art keywords
- layer
- signal sequence
- underground
- signal
- size
- 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.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
- G08—SIGNALLING
- G08B—SIGNALLING OR CALLING SYSTEMS; ORDER TELEGRAPHS; ALARM SYSTEMS
- G08B13/00—Burglar, theft or intruder alarms
- G08B13/16—Actuation by interference with mechanical vibrations in air or other fluid
- G08B13/1654—Actuation by interference with mechanical vibrations in air or other fluid using passive vibration detection systems
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Burglar Alarm Systems (AREA)
Abstract
The invention discloses an underground intrusion signal identification method in a field cultural relic protection system, and belongs to the field of deep learning and pattern identification. The method comprises the following steps: preprocessing the monitored underground intrusion signal, and using the preprocessed underground intrusion signal as an input sample of a convolutional neural network; respectively constructing a signal sample and a corresponding category thereof into training set data and test set data according to the category of the underground intrusion signal to be identified, and training the constructed neural network by using the constructed training set data; and according to the trained convolutional neural network, identifying each preprocessed test set data signal sample to obtain the category of the calculated signal, and comparing the category with a correct result. The underground intrusion signal classification recognition method based on the one-dimensional convolutional neural network classification model overcomes the defect that the traditional artificial monitoring is used for early warning, has universality and flexibility for underground intrusion signal recognition, and improves the recognition accuracy.
Description
Technical Field
The invention belongs to the technical field of deep learning and pattern recognition, and particularly relates to an underground intrusion signal recognition method and system in a field cultural relic protection system.
Background
Nowadays, effectively monitoring the field cultural relic stealing behavior is the key for protecting the field cultural relics all over the country, when the field cultural relics are stolen, illegal molecules often generate a single underground invasion signal, and an important means for effectively monitoring the field cultural relic stealing behavior is to perform real-time monitoring and identification on the underground invasion signal generated by the stealing behavior, so that the method is not only beneficial to protecting the field cultural relics all over the country from being influenced by stealing and resale, but also can prevent the country from huge economic and cultural losses, and has great significance.
The first method is the means of video monitoring, infrared monitoring technology and the like which are widely applied at present, and the condition of a field cultural relic protection site is artificially monitored by a monitoring camera; the second method is a time-frequency energy ratio (time-frequency energy ratio) based method for extracting time-frequency domain features, which introduces the extracted time-frequency features into algorithms such as Support Vector Machine (SVM) and Wavelet neural Network (Wavelet Network), and uses the calculated model to identify underground intrusion signals.
However, the existing underground intrusion monitoring methods all have some technical problems which are not negligible, and in the first method, the monitoring camera is easily affected by weather and installation environment, and meanwhile, monitoring depends on observation of operators on duty, so that the monitoring is limited to human self, and the monitoring is difficult to concentrate attention for a long time, thereby causing report omission; for the second method, in order to improve the identification accuracy, in the extraction process of the time-frequency features, time-domain segmentation windows and feature frequency band parameters need to be obtained through manual analysis, and the robustness is poor; in addition, wavelet neural networks are less versatile.
Disclosure of Invention
Aiming at the defects or improvement requirements in the prior art, the invention provides an underground intrusion signal identification method and an underground intrusion signal identification system in a field cultural relic protection system, and aims to solve the technical problems that missed reports and false reports are easy to occur in the existing field cultural relic protection method based on artificial monitoring, and the model robustness and universality are poor in the existing field cultural relic protection method based on algorithms such as machine learning and the like.
To achieve the above object, according to one aspect of the present invention, there is provided a method for identifying an underground intrusion signal in a field cultural relic protection system, comprising the steps of:
(1) acquiring an underground invasion signal from a field cultural relic protection system, judging whether the amplitude of the underground invasion signal is greater than a preset threshold value, if so, entering the step (2), otherwise, continuously acquiring the underground invasion signal, and repeating the step;
(2) acquiring an underground invasion signal within t time from the current time, and performing analog-to-digital conversion processing on the underground invasion signal by using an analog-to-digital converter to obtain a discrete underground invasion signal sequence;
(3) setting a counter i to be 1;
(4) judging whether the counter i is larger than the total length of the discrete underground invasion signal sequence obtained in the step (2), if so, ending the process, otherwise, turning to the step (5);
(5) judging whether the ith element of the discrete underground intrusion signal sequence obtained in the step (2) is greater than or equal to a preset threshold value, if so, entering the step (6), otherwise, entering the step (8);
(6) taking the first m elements and the last m elements of the ith element of the discrete underground invasion signal sequence to form a signal sequence, if the first m elements or the last m elements do not exist, executing zero filling operation to form a signal sequence with the final length of 2m +1, and carrying out zero filling or truncation processing on the obtained signal sequence to obtain a new signal sequence with the fixed length, wherein the value range of m is 1600-2400;
(7) inputting the signal sequence with the fixed length obtained in the step (6) into a trained one-dimensional convolutional neural network classification model, judging the type of the underground intrusion signal according to the output result of the one-dimensional convolutional neural network classification model, setting a counter i to be i + m, and returning to the step (4);
(8) and setting the counter i to i +1, and returning to the step (4).
Preferably, the value range of the preset threshold in the step (1) is 1 to 3.3 volts, and is preferably equal to 1.5 volts; the value range of t in the step (2) is 8 to 12 seconds, preferably 10 seconds; the value range of the preset threshold in the step (5) is 1000 to 1500, preferably 1200.
Preferably, the new signal sequence of fixed length in step (6) has a length of 3993; if the length of the signal sequence with the length of 2m +1 is larger than 3993, the step (6) is to apply truncation processing to the signal sequence, i.e., to truncate the element at the end of the signal sequence so that the length of the sequence is equal to 3993, and if the length of the signal sequence with the length of 2m +1 is smaller than 3993, the step (6) is to apply zero padding processing to the signal sequence, i.e., to zero pad the end of the signal sequence so that the length of the sequence is equal to 3993.
Preferably, in the step (7), if the output result of the one-dimensional convolutional neural network classification model is 0, it indicates that the underground intrusion signal is a walking signal, and if the output result is 1, it indicates that the underground intrusion signal is a mining signal.
Preferably, the one-dimensional convolutional neural network classification model is obtained by training through the following steps:
(1-1) acquiring an underground invasion signal within t time from the current time, and performing analog-to-digital conversion processing on the underground invasion signal by using an analog-to-digital converter to obtain a discrete underground invasion signal sequence;
(1-2) setting a counter j to 1;
(1-3) judging whether the counter j is larger than the length of the discrete underground invasion signal sequence, if so, entering the step (1-8), otherwise, entering the step (1-4);
(1-4) acquiring the jth element of the discrete underground intrusion signal sequence obtained in the step (1-1), judging whether the jth element is larger than or equal to a preset threshold value, if so, entering the step (1-5), otherwise, entering the step (1-6);
(1-5) taking the first n elements and the last n elements of the jth element from the discrete underground intrusion signal sequence to form a signal sequence, wherein if the first n elements or the last n elements do not exist, zero filling operation is executed to form a signal sequence with the final length of 2n +1, zero filling or truncation processing is carried out on the obtained signal sequence to obtain a new signal sequence with the fixed length, and then the step (1-7) is carried out, wherein the value range of n is 1600-2400;
(1-6) setting a counter j ═ j +1, and returning to the step (1-3);
(1-7) setting a counter j ═ j + n, and returning to the step (1-3);
(1-8) carrying out normalization processing on all the obtained new signal sequences with fixed lengths to obtain a plurality of normalized new signal sequences;
(1-9) dividing the plurality of normalized new signal sequences obtained in the step (1-8) into a training set and a testing set according to a ratio of 9:1, taking N data sets from all the training sets and inputting the data sets into a one-dimensional convolutional neural network classification model, and updating and optimizing weight parameters of each layer in the one-dimensional convolutional neural network classification model by using a back propagation algorithm to make loss values converge so as to obtain a trained one-dimensional convolutional neural network classification model; where N is a natural number with a minimum value of 16 and a maximum value of the total number of training sets.
Preferably, the normalization process of step (1-8) uses the following formula:
where x is an element in the new signal sequence of fixed length, xminIs the minimum value, x, in a new signal sequence of fixed lengthmaxIs the maximum value in the new signal sequence of fixed length.
Preferably, the one-dimensional convolutional neural network classification model includes an input layer, an output layer, 3 convolutional layers, 3 pooling layers, and 1 fully-connected layer, and its specific network structure is as follows:
the first layer is an input layer, and the input layer is constructed for N data sets, and the input size is N × 3993 × 1;
the second layer is a convolution layer, the input data with the size of N < x > 3993 < x > 1 uses 8 convolution kernels with the size of 1 < x > 3, the input data is convoluted by the step length of 3, all 0 is used for filling, the size of the obtained output data is N < x > 1331 < x > 8, and the output data is input to the third layer;
the third layer is a pooling layer, data with the input size of N × 1331 × 8 is subjected to maximum pooling operation with the step size of 2 by using the pooling size of 1 × 3, and the obtained output data with the size of N × 666 × 8 is input into the fourth layer;
the fourth layer is a convolution layer, the input data with the size of N666X 8 uses 16 convolution kernels with the size of 1X 3 to convolute the input data with the step length of 3, all 0 is used for filling, the obtained output data with the size of N222X 16 is input to the fifth layer;
the fifth layer is a pooling layer, data with the input size of N × 222 × 16 is subjected to maximum pooling operation with the step size of 2 by using a pooling size of 1 × 3, and output data with the size of N × 111 × 16 is obtained and input to the sixth layer;
the sixth layer is a convolution layer, data with the input size of N × 111 × 8 is convolved with input data with the step size of 3 by using 32 convolution kernels with the size of 1 × 3, all 0 padding is used, and the obtained output data with the size of N × 37 × 32 is input to the seventh layer;
the seventh layer is a pooling layer, data with the input size of N × 37 × 32 is subjected to maximum pooling operation with the step size of 2 by using a pooling size of 1 × 3, and output data with the size of N × 19 × 32 is obtained and input to the eighth layer;
the eighth layer is a fully-connected layer, the data input from the seventh layer is combined into one data with the size of 1 × 608, the layer has 128 neurons in total, each neuron is fully connected with each data in the data with the size of 1 × 608, and each neuron outputs one data, so that the layer outputs data with the size of N × 128 and inputs the data to the ninth layer.
The ninth layer is an output layer, which has 2 neurons in total, has an output data size of N × 2, and outputs the probability that the underground intrusion signal is a walking signal and the probability that the underground intrusion signal is a mining signal.
Preferably, when training the one-dimensional convolutional neural network classification model, the calculation formula of the output of the p-th convolutional layer is as follows:
wherein n ∈ [1, 3]]，
Representing the s characteristic vector of the p convolutional layer in the one-dimensional convolutional neural network classification model;
representing the q characteristic vector of the p-1 convolutional layer of the one-dimensional convolutional neural network classification model; denotes a one-dimensional convolution operation;
an s-th weight parameter representing a q-th feature vector in a p-th convolutional layer in the one-dimensional convolutional neural network classification model,
represents the s-th bias parameter of the p-th convolutional layer in the one-dimensional convolutional neural network classification model, and sigma (-) represents the activation function ReLu ().
Preferably, the parameters to be updated in each layer of the one-dimensional convolutional neural network classification model include a weight parameter and a bias parameter, an initial value of the weight parameter is a random value output by using a truncated normal distribution with a standard deviation of 0.1, and an initial value of the bias parameter is 0;
the loss value L of the one-dimensional convolutional neural network classification model is as follows:
where K denotes the number of classes, tl,kRepresenting the prediction result of the kth class after the I normalized new signal sequence is input into the one-dimensional convolution neural network classification model, yi,kRepresents the true result corresponding to the ith class i new signal sequence after normalization, K ∈ [1, K]And λ represents the degree of regularization, and the value of λ is 0.01, and W is a weight parameter of the one-dimensional convolutional neural network classification model, which is updated along with the training of the network.
According to another aspect of the present invention, there is provided an underground intrusion signal recognition system in a field cultural relic protection system, including:
the system comprises a first module, a second module and a duplication module, wherein the first module is used for acquiring an underground invasion signal from a field cultural relic protection system, judging whether the amplitude of the underground invasion signal is greater than a preset threshold value, if so, entering the second module, and if not, continuously acquiring the underground invasion signal and duplicating the module;
the second module is used for acquiring an underground invasion signal within the time t from the current time, and performing analog-to-digital conversion processing on the underground invasion signal by using an analog-to-digital converter to obtain a discrete underground invasion signal sequence;
a third module for setting a counter i equal to 1;
the fourth module is used for judging whether the counter i is larger than the total length of the discrete underground invasion signal sequence obtained by the second module, if so, the process is ended, otherwise, the fifth module is switched to;
the fifth module is used for judging whether the ith element of the discrete underground intrusion signal sequence obtained by the second module is greater than or equal to a preset threshold value, if so, entering the sixth module, and otherwise, entering the eighth module;
a sixth module, configured to take the first m elements and the last m elements of the ith element of the discrete underground intrusion signal sequence to form a signal sequence, where if the first m elements or the last m elements do not exist, a zero padding operation is performed to form a signal sequence with a final length of 2m +1, and zero padding or truncation processing is performed on the obtained signal sequence to obtain a new signal sequence with a fixed length, where a value range of m is 1600 to 2400;
the seventh module is used for inputting the signal sequence with the fixed length obtained by the sixth module into the trained one-dimensional convolutional neural network classification model, judging the type of the underground intrusion signal according to the output result of the one-dimensional convolutional neural network classification model, setting a counter i to be i + m, and returning the counter i to the fourth module;
and the eighth module is used for setting the counter i to i +1 and returning to the fourth module.
In general, compared with the prior art, the above technical solution contemplated by the present invention can achieve the following beneficial effects:
1. the method adopts the steps (1) to (8), the intrusion signals caused by the underground intrusion events are acquired by using acquisition equipment, the single intrusion sample data is segmented, and the one-dimensional convolutional neural network classification model is used as an identification classifier, so that the intrusion events can be effectively identified and classified by using the characteristic learning capacity of the convolutional neural network, and the technical problems of easy false alarm and missing report of the intrusion monitoring method in the field cultural relic protection system based on camera monitoring in the prior art are effectively solved.
2. Because the steps (6) and (7) are adopted, the invention carries out single underground intrusion signal sample segmentation on each group of sample data in the aspect of data processing, improves the universality of data processing, and is more intelligent, thereby effectively solving the technical problems of poor robustness and universality of the underground intrusion signal identification method in the existing field cultural relic protection system based on algorithms such as machine learning and the like.
Drawings
Fig. 1 is a flowchart of a method for identifying an underground intrusion signal in a field cultural relic protection system according to the invention.
FIG. 2 is a graph of loss values and classification accuracy used in the training process of the present invention.
Fig. 3 is a schematic structural diagram of a one-dimensional convolutional neural network classification model used in the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention 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 invention and are not intended to limit the invention. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
The invention provides an underground intrusion signal identification method in a field cultural relic protection system, which combines the idea of signal classification and the underground intrusion signal micro-vibration monitoring technology, wherein the underground intrusion signals generated by different trigger events propagated in soil have specific time-frequency characteristics. Therefore, the invention takes the trigger reason of the underground invasion signal as the mark of the signal and uses the one-dimensional convolutional neural network classification model as a classifier to predict.
As shown in fig. 1, the present invention provides a method for identifying an underground intrusion signal in a field cultural relic protection system, which comprises the following steps:
(1) acquiring an underground invasion signal from a field cultural relic protection system, judging whether the amplitude of the underground invasion signal is greater than a preset threshold value, if so, entering the step (2), otherwise, continuously acquiring the underground invasion signal, and repeating the step;
specifically, the underground invasion signal in the field cultural relic protection system is obtained through a micro-vibration sensor;
the value range of the preset threshold is 1 to 3.3 volts, preferably equal to 1.5 volts;
(2) acquiring an underground invasion signal within t time from the current time, and performing analog-to-digital conversion processing on the underground invasion signal by using an analog-to-digital converter to obtain a discrete underground invasion signal sequence;
specifically, t ranges from 8 to 12 seconds, preferably 10 seconds.
(3) Setting a counter i to be 1;
(4) judging whether the counter i is larger than the total length of the discrete underground invasion signal sequence obtained in the step (2), if so, ending the process, otherwise, turning to the step (5);
(5) judging whether the ith element of the discrete underground intrusion signal sequence obtained in the step (2) is greater than or equal to a preset threshold value, if so, entering the step (6), otherwise, entering the step (8);
specifically, the value of the preset threshold ranges from 1000 to 1500, preferably 1200.
(6) Taking the first m elements and the last m elements of the ith element of the discrete underground invasion signal sequence to form a signal sequence, if the first m elements or the last m elements do not exist, executing zero filling operation to form a signal sequence with the final length of 2m +1, and carrying out zero filling or truncation processing on the obtained signal sequence to obtain a new signal sequence with the fixed length;
specifically, m ranges from 1600 to 2400, preferably 2000.
The length of the new signal sequence with fixed length in this step is 3993.
If the length of the signal sequence with the length of 2m +1 is greater than 3993, truncation processing is applied to the signal sequence, i.e., elements at the end of the signal sequence are truncated so that the sequence length is equal to 3993, and if the length of the signal sequence with the length of 2m +1 is less than 3993, zero padding processing is applied to the signal sequence, i.e., zero padding is applied to the end of the signal sequence so that the sequence length is equal to 3993.
(7) Inputting the signal sequence with the fixed length obtained in the step (6) into a trained one-dimensional convolutional neural network classification model, judging the type of the underground intrusion signal according to the output result of the one-dimensional convolutional neural network classification model, setting a counter i to be i + m, and returning to the step (4);
specifically, if the output result of the one-dimensional convolutional neural network classification model is 0, it indicates that the underground intrusion signal is a walking signal, and if the output result is 1, it indicates that the underground intrusion signal is a mining signal.
(8) And setting the counter i to i +1, and returning to the step (4).
The one-dimensional convolutional neural network classification model is obtained by training through the following steps:
(1) acquiring an underground invasion signal within t time from the current time, and performing analog-to-digital conversion processing on the underground invasion signal by using an analog-to-digital converter to obtain a discrete underground invasion signal sequence;
(2) setting a counter j equal to 1;
(3) judging whether the counter j is larger than the length of the discrete underground intrusion signal sequence, if so, entering the step (8), and otherwise, entering the step (4);
(4) acquiring the jth element of the discrete underground intrusion signal sequence obtained in the step (1), judging whether the jth element is greater than or equal to a preset threshold value, if so, entering the step (5), otherwise, entering the step (6);
in particular, the value of the preset threshold ranges from 1000 to 1500, preferably equal to 1200.
(5) Taking the first n elements and the last n elements of the jth element from the discrete underground intrusion signal sequence to form a signal sequence, wherein if the first n elements or the last n elements do not exist, zero filling operation is executed to form a signal sequence with the final length of 2n +1, zero filling or truncation processing is carried out on the obtained signal sequence to obtain a new signal sequence with the fixed length, and then the step (7) is carried out;
specifically, n ranges from 1600 to 2400, preferably 2000.
The length of the new signal sequence with fixed length in this step is 3993.
(6) Setting a counter j ═ j +1, and returning to the step (3);
(7) setting a counter j ═ j + n, and returning to the step (3);
(8) carrying out normalization processing on all the obtained new signal sequences with fixed lengths to obtain a plurality of normalized new signal sequences;
specifically, the normalization process uses the following formula:
the purpose of normalization is to improve the model precision and accelerate the network learning speed. Where x is an element in the new signal sequence of fixed length, xminIs the minimum value, x, in a new signal sequence of fixed lengthmaxIs the maximum value in the new signal sequence of fixed length.
(9) Dividing the plurality of normalized new signal sequences obtained in the step (8) into a training set and a testing set according to a ratio of 9:1, taking N data sets from all the training sets and inputting the data sets into a one-dimensional convolutional neural network classification model, and updating and optimizing weight parameters of each layer in the one-dimensional convolutional neural network classification model by using a back propagation algorithm to make loss values converge so as to obtain a trained one-dimensional convolutional neural network classification model; where N is a natural number with a minimum value of 16 and a maximum value of the total number of training sets.
As shown in fig. 3, for the one-dimensional convolutional neural network classification model in the present invention, the one-dimensional convolutional neural network classification model includes an input layer, an output layer, 3 one-dimensional convolutional layers, 3 one-dimensional pooling layers, and 1 full-link layer, and its specific network structure is as follows:
the first layer is an input layer, and the input layer is constructed for N data sets, and the input size is N × 3993 × 1;
the second layer is a convolution layer, the input data with the size of N < x > 3993 < x > 1 uses 8 convolution kernels with the size of 1 < x > 3, the input data is convoluted by the step length of 3, all 0 is used for filling, the size of the obtained output data is N < x > 1331 < x > 8, and the output data is input to the third layer;
the third layer is a pooling layer, data with the input size of N × 1331 × 8 is subjected to maximum pooling operation with the step size of 2 by using the pooling size of 1 × 3, and the obtained output data with the size of N × 666 × 8 is input into the fourth layer;
the fourth layer is a convolution layer, the input data with the size of N666X 8 uses 16 convolution kernels with the size of 1X 3 to convolute the input data with the step length of 3, all 0 is used for filling, the obtained output data with the size of N222X 16 is input to the fifth layer;
the fifth layer is a pooling layer, data with the input size of N × 222 × 16 is subjected to maximum pooling operation with the step size of 2 by using a pooling size of 1 × 3, and output data with the size of N × 111 × 16 is obtained and input to the sixth layer;
the sixth layer is a convolution layer, data with the input size of N × 111 × 8 is convolved with input data with the step size of 3 by using 32 convolution kernels with the size of 1 × 3, all 0 padding is used, and the obtained output data with the size of N × 37 × 32 is input to the seventh layer;
the seventh layer is a pooling layer, data with the input size of N × 37 × 32 is subjected to maximum pooling operation with the step size of 2 by using a pooling size of 1 × 3, and output data with the size of N × 19 × 32 is obtained and input to the eighth layer;
the eighth layer is a fully-connected layer, the data input from the seventh layer is combined into one data (one-dimensional data) with the size of 1 × 608, the layer has 128 neurons in total, each neuron is fully connected with each data in the data with the size of 1 × 608, and each neuron outputs one data, so that the layer outputs data with the size of N × 128 and inputs the data to the ninth layer.
The ninth layer is an output layer, which has 2 neurons in total, has an output data size of N × 2, and outputs the probability that the underground intrusion signal is a walking signal and the probability that the underground intrusion signal is a mining signal.
Specifically, when training the one-dimensional convolutional neural network classification model, the computation formula of the output quantity of the p-th convolutional layer (where p ∈ [1, 3]) is:
wherein,
representing the s characteristic vector of the p convolutional layer in the one-dimensional convolutional neural network classification model;
representing the q characteristic vector of the p-1 convolutional layer of the one-dimensional convolutional neural network classification model; denotes a one-dimensional convolution operation;
an s-th weight parameter representing a q-th feature vector in a p-th convolutional layer in the one-dimensional convolutional neural network classification model,
represents the s-th bias parameter of the p-th convolutional layer in the one-dimensional convolutional neural network classification model, and sigma (-) represents the activation function ReLu ().
The parameters to be updated of each layer in the one-dimensional convolutional neural network classification model comprise a weight parameter and a bias parameter, the initial value of the weight parameter is a random value output by using a truncation type normal distribution with a standard deviation of 0.1, and the initial value of the bias parameter is 0.
Wherein, the loss value L of the one-dimensional convolution neural network classification model is as follows:
where K represents the number of classes (in the present embodiment, it is 2), tl,kRepresenting the prediction result of the kth class after the I normalized new signal sequence is input into the one-dimensional convolution neural network classification model, yi,kRepresents the true result corresponding to the ith class i new signal sequence after normalization, K ∈ [1, K]And λ represents the degree of regularization, and the value of λ is 0.01, and W is a weight parameter of the one-dimensional convolutional neural network classification model, which is updated along with the training of the network.
In order to accelerate the reduction speed of the loss value, the invention adopts a Batch gradient descent (Batch gradient decision) algorithm as an optimizer to process the output result of the one-dimensional convolutional neural network classification model, the processed result is used as the input of the one-dimensional convolutional neural network classification model, and the output result of the one-dimensional convolutional neural network classification model is obtained, …, until the loss value of the whole network is converged.
The objective of using the batch gradient descent algorithm is to ensure convergence to the extreme point because the algorithm converges in the correct direction every time it is updated.
The loss value and classification accuracy rate curve used in the training process of the invention is shown in fig. 2, and the final test set data can be found to have 96% classification accuracy rate according to the change curve.
Test results
148 test set samples are input into the one-dimensional convolutional neural network classification model trained by the method, and the network automatically identifies the underground invasion signal type and gives an identification result. As shown in table 1 below, it can be clearly seen that the identification accuracy of the present invention is higher compared to the two underground intrusion signal identification methods used in the art mentioned in the background of the present invention.
TABLE 1
| Method of producing a composite material | Rate of identification accuracy |
| Intrusion event identification method based on time-frequency energy ratio characteristics | 94％ |
| Richer wavelet neural network-based intrusion event identification method | 95％ |
| The invention | 96％ |
It will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the invention, and that any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the present invention.
Claims (10)
1. An underground intrusion signal identification method in a field cultural relic protection system is characterized by comprising the following steps:
(1) acquiring an underground invasion signal from a field cultural relic protection system, judging whether the amplitude of the underground invasion signal is greater than a preset threshold value, if so, entering the step (2), otherwise, continuously acquiring the underground invasion signal, and repeating the step;
(2) acquiring an underground invasion signal within t time from the current time, and performing analog-to-digital conversion processing on the underground invasion signal by using an analog-to-digital converter to obtain a discrete underground invasion signal sequence;
(3) setting a counter i to be 1;
(4) judging whether the counter i is larger than the total length of the discrete underground invasion signal sequence obtained in the step (2), if so, ending the process, otherwise, turning to the step (5);
(5) judging whether the ith element of the discrete underground intrusion signal sequence obtained in the step (2) is greater than or equal to a preset threshold value, if so, entering the step (6), otherwise, entering the step (8);
(6) taking the first m elements and the last m elements of the ith element of the discrete underground invasion signal sequence to form a signal sequence, if the first m elements or the last m elements do not exist, executing zero filling operation to form a signal sequence with the final length of 2m +1, and carrying out zero filling or truncation processing on the obtained signal sequence to obtain a new signal sequence with the fixed length, wherein the value range of m is 1600-2400;
(7) inputting the signal sequence with the fixed length obtained in the step (6) into a trained one-dimensional convolutional neural network classification model, judging the type of the underground intrusion signal according to the output result of the one-dimensional convolutional neural network classification model, setting a counter i to be i + m, and returning to the step (4);
(8) and setting the counter i to i +1, and returning to the step (4).
2. A subterranean intrusion signal identification method according to claim 1,
the value range of the preset threshold in the step (1) is 1 to 3.3 volts;
the value range of t in the step (2) is 8 to 12 seconds;
the value range of the preset threshold in the step (5) is 1000 to 1500.
3. A subterranean intrusion signal identification method according to claim 1,
the length of the new signal sequence with fixed length in the step (6) is 3993;
if the length of the signal sequence with the length of 2m +1 is larger than 3993, the step (6) is to apply truncation processing to the signal sequence, i.e., to truncate the element at the end of the signal sequence so that the length of the sequence is equal to 3993, and if the length of the signal sequence with the length of 2m +1 is smaller than 3993, the step (6) is to apply zero padding processing to the signal sequence, i.e., to zero pad the end of the signal sequence so that the length of the sequence is equal to 3993.
4. The method according to claim 1, wherein in the step (7), if the output result of the one-dimensional convolutional neural network classification model is 0, it indicates that the underground intrusion signal is a walk signal, and if the output result is 1, it indicates that the underground intrusion signal is a mining signal.
5. The method of claim 1, wherein the one-dimensional convolutional neural network classification model is trained by the following steps:
(1-1) acquiring an underground invasion signal within t time from the current time, and performing analog-to-digital conversion processing on the underground invasion signal by using an analog-to-digital converter to obtain a discrete underground invasion signal sequence;
(1-2) setting a counter j to 1;
(1-3) judging whether the counter j is larger than the length of the discrete underground invasion signal sequence, if so, entering the step (1-8), otherwise, entering the step (1-4);
(1-4) acquiring the jth element of the discrete underground intrusion signal sequence obtained in the step (1-1), judging whether the jth element is larger than or equal to a preset threshold value, if so, entering the step (1-5), otherwise, entering the step (1-6);
(1-5) taking the first n elements and the last n elements of the jth element from the discrete underground intrusion signal sequence to form a signal sequence, wherein if the first n elements or the last n elements do not exist, zero filling operation is executed to form a signal sequence with the final length of 2n +1, zero filling or truncation processing is carried out on the obtained signal sequence to obtain a new signal sequence with the fixed length, and then the step (1-7) is carried out, wherein the value range of n is 1600-2400;
(1-6) setting a counter j ═ j +1, and returning to the step (1-3);
(1-7) setting a counter j ═ j + n, and returning to the step (1-3);
(1-8) carrying out normalization processing on all the obtained new signal sequences with fixed lengths to obtain a plurality of normalized new signal sequences;
(1-9) dividing the plurality of normalized new signal sequences obtained in the step (1-8) into a training set and a testing set according to a ratio of 9:1, taking N data sets from all the training sets and inputting the data sets into a one-dimensional convolutional neural network classification model, and updating and optimizing weight parameters of each layer in the one-dimensional convolutional neural network classification model by using a back propagation algorithm to make loss values converge so as to obtain a trained one-dimensional convolutional neural network classification model; where N is a natural number with a minimum value of 16 and a maximum value of the total number of training sets.
6. The method of claim 5, wherein the normalization in step (1-8) is performed using the following equation:
where x is an element in the new signal sequence of fixed length, xminIs the minimum value, x, in a new signal sequence of fixed lengthmaxIs the maximum value in the new signal sequence of fixed length.
7. A subterranean intrusion signal identification method according to claim 5 or 6,
the one-dimensional convolutional neural network classification model comprises an input layer, an output layer, 3 convolutional layers, 3 pooling layers and 1 full-connection layer, and the specific network structure is as follows:
the first layer is an input layer, and the input layer is constructed for N data sets, and the input size is N × 3993 × 1;
the second layer is a convolution layer, the input data with the size of N < x > 3993 < x > 1 uses 8 convolution kernels with the size of 1 < x > 3, the input data is convoluted by the step length of 3, all 0 is used for filling, the size of the obtained output data is N < x > 1331 < x > 8, and the output data is input to the third layer;
the third layer is a pooling layer, data with the input size of N × 1331 × 8 is subjected to maximum pooling operation with the step size of 2 by using the pooling size of 1 × 3, and the obtained output data with the size of N × 666 × 8 is input into the fourth layer;
the fourth layer is a convolution layer, the input data with the size of N666X 8 uses 16 convolution kernels with the size of 1X 3 to convolute the input data with the step length of 3, all 0 is used for filling, the obtained output data with the size of N222X 16 is input to the fifth layer;
the fifth layer is a pooling layer, data with the input size of N × 222 × 16 is subjected to maximum pooling operation with the step size of 2 by using a pooling size of 1 × 3, and output data with the size of N × 111 × 16 is obtained and input to the sixth layer;
the sixth layer is a convolution layer, data with the input size of N × 111 × 8 is convolved with input data with the step size of 3 by using 32 convolution kernels with the size of 1 × 3, all 0 padding is used, and the obtained output data with the size of N × 37 × 32 is input to the seventh layer;
the seventh layer is a pooling layer, data with the input size of N × 37 × 32 is subjected to maximum pooling operation with the step size of 2 by using a pooling size of 1 × 3, and output data with the size of N × 19 × 32 is obtained and input to the eighth layer;
the eighth layer is a fully-connected layer, the data input by the seventh layer is combined into data with the size of 1 × 608, the layer has 128 neurons, each neuron is fully connected with each data in the data with the size of 1 × 608, each neuron outputs one data, and thus the layer outputs data with the size of N × 128 and inputs the data to the ninth layer;
the ninth layer is an output layer, which has 2 neurons in total, has an output data size of N × 2, and outputs the probability that the underground intrusion signal is a walking signal and the probability that the underground intrusion signal is a mining signal.
8. The method of claim 7, wherein when training the one-dimensional convolutional neural network classification model, the p convolutional layer output is calculated as:
wherein n ∈ [1, 3]]，
Representing the s characteristic vector of the p convolutional layer in the one-dimensional convolutional neural network classification model;
representing the q characteristic vector of the p-1 convolutional layer of the one-dimensional convolutional neural network classification model; denotes a one-dimensional convolution operation;
an s-th weight parameter representing a q-th feature vector in a p-th convolutional layer in the one-dimensional convolutional neural network classification model,
represents the s-th bias parameter of the p-th convolutional layer in the one-dimensional convolutional neural network classification model, and sigma (-) represents the activation function ReLu ().
9. A subterranean intrusion signal identification method according to claim 7,
parameters needing to be updated in each layer in the one-dimensional convolutional neural network classification model comprise a weight parameter and a bias parameter, the initial value of the weight parameter is a random value output by using a truncation type normal distribution with a standard deviation of 0.1, and the initial value of the bias parameter is 0;
the loss value L of the one-dimensional convolutional neural network classification model is as follows:
where K denotes the number of classes, tl,kRepresenting the prediction result of the kth class after the I normalized new signal sequence is input into the one-dimensional convolution neural network classification model, yi,kRepresents the true result corresponding to the ith class i new signal sequence after normalization, K ∈ [1, K]λ represents the degree of regularization, which is 0.01, and W is a one-dimensional convolutionAnd the weight parameters of the neural network classification model are updated along with the training of the network.
10. An underground intrusion signal recognition system in a field cultural relic protection system, comprising:
the system comprises a first module, a second module and a duplication module, wherein the first module is used for acquiring an underground invasion signal from a field cultural relic protection system, judging whether the amplitude of the underground invasion signal is greater than a preset threshold value, if so, entering the second module, and if not, continuously acquiring the underground invasion signal and duplicating the module;
the second module is used for acquiring an underground invasion signal within the time t from the current time, and performing analog-to-digital conversion processing on the underground invasion signal by using an analog-to-digital converter to obtain a discrete underground invasion signal sequence;
a third module for setting a counter i equal to 1;
the fourth module is used for judging whether the counter i is larger than the total length of the discrete underground invasion signal sequence obtained by the second module, if so, the process is ended, otherwise, the fifth module is switched to;
the fifth module is used for judging whether the ith element of the discrete underground intrusion signal sequence obtained by the second module is greater than or equal to a preset threshold value, if so, entering the sixth module, and otherwise, entering the eighth module;
a sixth module, configured to take the first m elements and the last m elements of the ith element of the discrete underground intrusion signal sequence to form a signal sequence, where if the first m elements or the last m elements do not exist, a zero padding operation is performed to form a signal sequence with a final length of 2m +1, and zero padding or truncation processing is performed on the obtained signal sequence to obtain a new signal sequence with a fixed length, where a value range of m is 1600 to 2400;
the seventh module is used for inputting the signal sequence with the fixed length obtained by the sixth module into the trained one-dimensional convolutional neural network classification model, judging the type of the underground intrusion signal according to the output result of the one-dimensional convolutional neural network classification model, setting a counter i to be i + m, and returning the counter i to the fourth module;
and the eighth module is used for setting the counter i to i +1 and returning to the fourth module.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010029383.0A CN111242211B (en) | 2020-01-13 | 2020-01-13 | Underground intrusion signal identification method and system in field cultural relic protection system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010029383.0A CN111242211B (en) | 2020-01-13 | 2020-01-13 | Underground intrusion signal identification method and system in field cultural relic protection system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111242211A CN111242211A (en) | 2020-06-05 |
| CN111242211B true CN111242211B (en) | 2020-10-09 |
Family
ID=70874470
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010029383.0A Active CN111242211B (en) | 2020-01-13 | 2020-01-13 | Underground intrusion signal identification method and system in field cultural relic protection system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111242211B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115099636B (en) * | 2022-06-27 | 2023-10-31 | 中南民族大学 | Effectiveness evaluation method and system of cultural relics security system based on complex network theory |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103903374A (en) * | 2014-03-31 | 2014-07-02 | 北京工业大学 | Anti-theft device for an ancient tomb |
| CN104200592A (en) * | 2014-09-25 | 2014-12-10 | 北京世纪之星应用技术研究中心 | Perimeter protection alarm system utilizing linear displacement to detect invasion and linear displacement detector |
| CN204206232U (en) * | 2014-12-04 | 2015-03-11 | 西安工业大学 | A field intelligent cultural relic monitoring system based on vibration sensor |
| CN105954791A (en) * | 2016-06-01 | 2016-09-21 | 长江大学 | Vibration ground wave fiber sensing detection system for prevention of burglary and excavation of underground historical relics |
| CN106127135A (en) * | 2016-06-21 | 2016-11-16 | 长江大学 | A kind of Ling Qu invasion vibration signal characteristics extracts and classification and identification algorithm |
| CN107086869A (en) * | 2017-06-09 | 2017-08-22 | 武汉旗云高科工程技术有限公司 | A kind of underground invasion signal recognition method encoded based on arrowband and system |
| CN110415471A (en) * | 2019-07-03 | 2019-11-05 | 浙江大学 | A 6LowPan-based anti-theft monitoring terminal and method for underground cultural relics |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7085749B2 (en) * | 2001-05-31 | 2006-08-01 | Canon Kabushiki Kaisha | Pulse signal circuit, parallel processing circuit, pattern recognition system, and image input system |
-
2020
- 2020-01-13 CN CN202010029383.0A patent/CN111242211B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103903374A (en) * | 2014-03-31 | 2014-07-02 | 北京工业大学 | Anti-theft device for an ancient tomb |
| CN104200592A (en) * | 2014-09-25 | 2014-12-10 | 北京世纪之星应用技术研究中心 | Perimeter protection alarm system utilizing linear displacement to detect invasion and linear displacement detector |
| CN204206232U (en) * | 2014-12-04 | 2015-03-11 | 西安工业大学 | A field intelligent cultural relic monitoring system based on vibration sensor |
| CN105954791A (en) * | 2016-06-01 | 2016-09-21 | 长江大学 | Vibration ground wave fiber sensing detection system for prevention of burglary and excavation of underground historical relics |
| CN106127135A (en) * | 2016-06-21 | 2016-11-16 | 长江大学 | A kind of Ling Qu invasion vibration signal characteristics extracts and classification and identification algorithm |
| CN107086869A (en) * | 2017-06-09 | 2017-08-22 | 武汉旗云高科工程技术有限公司 | A kind of underground invasion signal recognition method encoded based on arrowband and system |
| CN110415471A (en) * | 2019-07-03 | 2019-11-05 | 浙江大学 | A 6LowPan-based anti-theft monitoring terminal and method for underground cultural relics |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111242211A (en) | 2020-06-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114755745B (en) | Hail weather recognition and classification method based on multi-channel deep residual shrinkage network | |
| CN109884419B (en) | An online fault diagnosis method for power quality of smart grid | |
| CN110213222A (en) | Network inbreak detection method based on machine learning | |
| CN113571133B (en) | Lactic acid bacteria antibacterial peptide prediction method based on graph neural network | |
| CN102841131B (en) | Intelligent steel cord conveyer belt defect identification method and intelligent steel cord conveyer belt defect identification system | |
| CN112560806B (en) | A kind of artificial intelligence identification method of natural gas pipeline leakage signal | |
| CN111898447A (en) | Fault feature extraction method for wind turbines based on symplectic geometric modal decomposition | |
| CN112949391B (en) | Intelligent security inspection method based on deep learning harmonic signal analysis | |
| CN108171119B (en) | SAR image change detection method based on residual network | |
| CN109239669B (en) | A self-evolving radar target detection algorithm based on deep learning | |
| CN110458219B (en) | phi-OTDR vibration signal identification algorithm based on STFT-CNN-RVFL | |
| CN111553186A (en) | Electromagnetic signal identification method based on depth long-time and short-time memory network | |
| CN109581339A (en) | A kind of sonar recognition methods based on brainstorming adjust automatically autoencoder network | |
| CN112766603A (en) | Traffic flow prediction method, system, computer device and storage medium | |
| CN114973019B (en) | A method and system for detecting and classifying geospatial information changes based on deep learning | |
| CN114252706B (en) | Lightning early warning method and system | |
| CN116993060A (en) | Intelligent building site safety management method and system based on Internet of things | |
| CN111476102A (en) | A security protection method, central control device and computer storage medium | |
| CN111242211B (en) | Underground intrusion signal identification method and system in field cultural relic protection system | |
| Wang et al. | A novel underground pipeline surveillance system based on hybrid acoustic features | |
| CN112613749A (en) | Cross-border hidden high-risk factor risk intelligent analysis system | |
| CN114266271B (en) | Distributed optical fiber vibration signal mode classification method and system based on neural network | |
| CN109615027B (en) | An intelligent prediction method for wind speed feature extraction along high-speed railway | |
| Zhou et al. | A specific emitter identification method based on RF-DNA and XGBoost | |
| Yang et al. | A structure optimization algorithm of neural networks for large-scale data sets |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |