CN-122017767-A - Coal mine radar signal abnormal body identification method based on improved generation countermeasure network

CN122017767ACN 122017767 ACN122017767 ACN 122017767ACN-122017767-A

Abstract

The invention relates to the technical field of coal mine safety and discloses a coal mine radar signal abnormal body identification method based on an improved generation countermeasure network, which comprises the following steps of S1, acquiring radar signals covering geology in a mine in real time, and constructing a multi-frequency-band radar signal set; the method comprises the steps of S2, preprocessing a multi-band radar signal set, S3, constructing a double-layer generation countermeasure network model which is introduced into a structural feature extraction module and a double-attention module, generating abnormal body feature information data, S4, generating abnormal report data based on the abnormal body feature information data, wherein the abnormal report data comprises data of positions, types and properties of abnormal bodies.

Inventors

CHEN HUITING
YU XUELEI
Jin Weihu
WANG HAOXING
Pan Yanqi

Assignees

淮北矿业股份有限公司

Dates

Publication Date: 20260512
Application Date: 20251219

Claims (10)

1. The coal mine radar signal abnormal body identification method based on the improved generation countermeasure network is characterized by comprising the following steps of: s1, acquiring radar signals covering geology in a mine in real time, and constructing a multi-band radar signal set; s2, preprocessing a multi-band radar signal set; S3, constructing a double-layer generation countermeasure network model which is introduced into the structural feature extraction module and the dual-attention module, processing the preprocessed multi-band radar signal set through the double-layer generation countermeasure network model, and generating abnormal body feature information data; And S4, generating abnormality report data based on the abnormality characteristic information data, wherein the abnormality report data comprises data of the position, the type and the property of the abnormality.
2. The method for identifying anomalies in radar signals in a coal mine based on an improved generation countermeasure network of claim 1, wherein step S1 includes the specific steps of: Step S11, deploying multi-band radar equipment at a plurality of positions in a mine, acquiring multi-band radar signals by using the multi-band radar equipment, setting acquisition time as T, and sampling frequency in the time T Signal acquisition is carried out on different positions in the mine to form a multi-band radar signal set Where n represents the number of radar signals acquired, m represents the number of acquisition points in the mine, Representing an ith multi-band radar signal acquired at a jth acquisition point; step S12, multi-band radar signals aiming at different acquisition points Recording the frequency distribution of a signal , wherein, Representing frequency, constructing a frequency distribution set of a multi-band radar signal ; Step S13, collecting multi-band radar signal set Frequency distribution set Forming a final multi-band radar signal set , 。
3. The method for identifying anomalies in radar signals in a coal mine based on an improved generation countermeasure network of claim 2, wherein step S2 includes the specific steps of: Step S21, for final multi-band radar signal set Denoising, namely denoising the multi-band radar signal set by adopting a self-adaptive denoising algorithm The noise in the radar signal is identified and removed, and the denoised multi-band radar signal set is recorded as ; Step S22, denoising the multi-band radar signal set Normalization processing is carried out, the amplitude range of the signal is scaled to the [0,1] interval by using a linear normalization method, and the normalized multi-band radar signal set is recorded as ; Step S23, normalizing the multiband radar signal Performing multistage filtering, and removing low-frequency and high-frequency interference components by using a band-pass filter to obtain a multi-band radar signal intermediate frequency signal Removing residual noise and uncorrelated signal interference by using an adaptive filter to obtain a preprocessed multi-band radar signal set 。
4. The method for recognizing abnormal bodies of radar signals of coal mine based on the improved generation of the countermeasure network according to claim 3, wherein the two-layer generation countermeasure network model in the step S3 includes a first-layer generation countermeasure network and a second-layer generation countermeasure network; the first layer generates an countermeasure network for extracting the preprocessed multi-band radar signal set In (2) generating a set of global feature vectors based on the global features And collect global feature vectors Input to the second layer generates an antagonism network, where p represents the number of eigenvectors, Representing a kth global feature vector; The second layer generates an countermeasure network for the global feature vector set Refinement and separation processing to generate an outlier feature vector set Wherein q represents the number of outlier feature vectors, Representing the first outlier feature vector.
5. The method for identifying anomalies in radar signals of a coal mine based on an improved generation countermeasure network of claim 4, wherein the first-tier generation countermeasure network includes a first-tier generator First layer discriminator Embedding first layer generators Encoder section and first layer arbiter of (c) A dual attention module in the feature extraction layer of (a), the dual attention module comprising a time-sequential attention module and a frequency-domain attention module; the calculation formula of the time sequence attention module is as follows: ; Wherein, the Representing the query matrix and, The matrix of keys is represented and, A matrix of values is represented and, Is a key vector Is used in the manufacture of a printed circuit board, Is a weighted feature representation of the attention weight; The calculation formula of the frequency domain attention module is as follows: ; Wherein, the Representing a preprocessed set of multi-band radar signals , Representing a set of preprocessed multi-band radar signals A fast fourier transform is performed and the data is processed, A multi-layer perceptron is shown, Representation of The function of the function is that, The product of the Hadamard is represented, Representing the spectral features after frequency domain weight enhancement.
6. The method for identifying anomalies in radar signals of a coal mine based on an improved generation countermeasure network of claim 5, wherein the second-tier generation countermeasure network includes a second-tier generator Second layer discriminator Integrated in a second layer generator Second layer discriminator Internal hierarchical attention refinement module and introduction of a second tier generator The hierarchical attention refining module comprises a feature selection attention network and an abnormal decoupling attention network which are connected in series; The calculation formula of the feature selection attention network is as follows: ; ; Wherein, the The attention weight vector is represented as such, And Is a learnable weight matrix and bias vector, A linear transformation is represented and is used to represent, Representation of The function of the function is that, The product of the Hadamard is represented, Representing passing weights Modulated features; The calculation formula of the abnormal decoupling attention network is as follows: ; ; Wherein, the Representing a query matrix, a key matrix and a value matrix respectively, Respectively representing the projection matrix of the i-th head, Representing a learnable output projection matrix; the calculation formula of the structural feature extraction module is as follows: ; Wherein, the Representing hidden code c and generating features The mutual information between the two pieces of information, Indicating the desired log-likelihood that it is likely, Is approximately true posterior distribution Is used for the neural network of (a), The entropy of the hidden code c is represented, Representing the hyper-parameters controlling the weights of the mutual information items.
7. The method for identifying anomalies in radar signals of a coal mine based on an improved generation countermeasure network of claim 6, wherein the second-layer generation countermeasure network output includes a set of decoupled and categorized anomaly characteristic vectors And its corresponding class label derived from the latent code c 。
8. The method for identifying anomalies in radar signals of a coal mine based on an improved generation countermeasure network of claim 7, further comprising: distinguishing device for generating countermeasure network in second layer An interpretability analysis module is then integrated to generate structured interpretation data for each identified outlier feature.
9. The method for recognizing abnormal bodies of radar signals in coal mine based on improved generation of countermeasure network according to claim 8, wherein said arbiter for generating countermeasure network in the second layer Then, integrating an interpretability analysis module, which specifically comprises the following steps: step 100, for a discriminator Feature map of last convolutional layer output of (2) Calculate its gradient to the final classification decision ; Step 200, obtaining weights through gradient global average pooling And generates class activation heatmaps ; ; ; Wherein Z is a feature map The total number of positions in the time dimension, Is the score of category c For characteristic diagram The gradient at the i-th position, Representing the overall importance weight of the device, Representation of A function; Step 300, class activation heat map And the preprocessed multi-band radar signal set Alignment in time/frequency domain; step 400, calculating the statistical characteristics of the original signals corresponding to the salient regions in the heat map, and generating a structural interpretation report comprising anomaly localization, anomaly type confidence level, key evidence characteristics and characteristic contribution level.
10. The method for identifying anomalies in radar signals in a coal mine based on an improved generation countermeasure network of claim 9, further comprising the step of: In the training process of generating the countermeasure network in a double-layer manner, the network structures of the generator and the discriminator are adjusted layer by layer, and the calculation mode of mutual information is optimized.

Description

Coal mine radar signal abnormal body identification method based on improved generation countermeasure network Technical Field The invention relates to the technical field of coal mine safety, in particular to a coal mine radar signal abnormal body identification method based on improved generation of an countermeasure network. Background Along with the continuous increase of the mining depth of the coal mine, the mine environment is increasingly complex, and safety monitoring and disaster early warning of the mine become important links for guaranteeing the safety production of the coal mine. Radar signal technology is an important mine monitoring means, and is widely applied to geological detection and safety monitoring of coal mines because of its ability to penetrate geological layers and provide high-precision underground structure information. However, the coal mine environment is complex and changeable, and the radar signal is influenced by various factors such as geological conditions, equipment noise, electromagnetic interference and the like in the propagation process, so that the signal quality is reduced, and the recognition accuracy of abnormal bodies is further influenced. The prior art has the following disadvantages: Firstly, the existing coal mine radar signal abnormal body identification method mainly depends on the traditional signal processing and statistical analysis technology. These methods typically identify potential anomalies by preprocessing the radar signal, feature extraction, and rule or model based anomaly detection. However, due to the specificity of the coal mine environment, the conventional method has a plurality of defects in processing complex and changeable signals. For example, it is difficult for conventional filtering and denoising methods to effectively eliminate non-gaussian noise and strong interference signals frequently occurring in coal mine environments, resulting in poor quality of the processed signals. Secondly, the anomaly detection method based on rules or models depends on preset thresholds or hypothesis models, and the thresholds or models can not effectively adapt to complex changes of coal mine environments in practical application, so that missed detection or false detection of anomalies is caused. Secondly, in the prior art, an abnormality detection method based on deep learning has been applied in some fields such as image recognition and voice recognition. However, the direct application of these methods to the processing of coal mine radar signals and anomaly identification remains a challenge. Firstly, the data volume of the radar signal of the coal mine is huge and the characteristics are complex, and when the data are processed by the existing deep learning models such as a Convolutional Neural Network (CNN) or a cyclic neural network (RNN), the over-fitting phenomenon is easy to occur, so that the generalization capability of the model is insufficient. In addition, most of the existing deep learning models focus on global feature extraction, but the recognition effect on the hidden local abnormal features in radar signals is poor, so that the recognition accuracy on abnormal bodies is not high. Thirdly, generating an countermeasure network as an important branch of deep learning is attracting attention in the field of anomaly detection gradually due to its advantages in data generation and feature extraction. The GAN enables the generator to generate samples close to the real data distribution through the countermeasure training of the generator and the arbiter, thereby exhibiting unique advantages in terms of data enhancement, anomaly detection, and the like. However, conventional GAN structures still face a number of problems when applied to coal mine radar signal processing. For example, when a single-layer GAN model processes high-dimensional, complex data, the training process of its generator and arbiter tends to be in an unstable state, resulting in poor quality of the resulting results. In addition, the feature extraction capability of the conventional GAN is mainly focused on global features, but the extraction and separation capability of local features is weak, so that it is difficult to effectively identify abnormal bodies in radar signals. Therefore, aiming at the defects and shortcomings of the prior art of coal mine radar signal processing and abnormal body recognition, a new technical method is needed to be better adapted to the complexity of a coal mine environment, and the accuracy and the robustness of radar signal abnormal body recognition are improved. Disclosure of Invention The invention aims to solve the problems and provide a coal mine radar signal abnormal body identification method based on improved generation of an countermeasure network. The invention provides a coal mine radar signal abnormal body identification method based on an improved generation countermeasure network, which comprises the following steps: s1