Deep learning-based abrupt change signal identification method and device, medium and equipment
Through the deep learning-based signal segmentation and feature scoring method, the problem that traditional methods are difficult to identify complex mutation signals is solved, and high-precision and real-time mutation signal recognition is achieved, adapted to multi-channel signal processing, and improved recognition accuracy and response capabilities.
Patent Information
- Application Number
- CN202510296335.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-13
- Publication Date
- 2025-07-08
AI Technical Summary
In explosion tests, traditional feature extraction methods are difficult to effectively capture complex mutant signal characteristics, such as high noise and strong mutational signals, resulting in difficulty in identification.
Using a deep learning-based method, we use the original signal, segmented processing, extract time-frequency features, score using the inter-class variance ratio method and select high-score features, and input it into the mutation signal recognition model for identification.
It improves the recognition accuracy and real-time of mutation signals, can quickly respond and accurately identify mutation signals in complex environments, adapt to multi-channel signal processing, reduce interference, and has good robustness and reliability.
Smart Images
Figure CN120277385A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of signal processing, and in particular, to a method and device, medium, and equipment for identifying mutation signals based on deep learning. Background Art
[0002] In recent years, intelligent signal processing methods based on deep learning have been widely applied. Although certain achievements have been made in signal recognition in existing research, there are still some challenges in processing mutation signals in explosion tests. For example, the signal features are complex, and the signals in explosion tests have characteristics such as high noise, strong mutation, and obvious time-variation. Traditional feature extraction methods are difficult to effectively capture these complex features. Summary of the Invention
[0003] In view of at least one of the above technical problems, embodiments of the present invention provide a method and device, medium, and equipment for identifying mutation signals based on deep learning.
[0004] According to a first aspect, the method for identifying mutation signals based on deep learning provided by embodiments of the present invention includes:
[0005] Collecting an original signal;
[0006] Segmenting the original signal at a preset time interval to obtain a plurality of signal segments;
[0007] Extracting time-frequency features from each signal segment;
[0008] Using the between-class variance ratio method to determine the score of each time-frequency feature of each signal segment; wherein, the score of each time-frequency feature is used to characterize the feature discrimination degree corresponding to the time-frequency feature;
[0009] Selecting time-frequency features with scores higher than a preset score from each time-frequency feature of each signal segment, and using the selected time-frequency features as the high-score features of the signal segment;
[0010] Inputting the high-score features of each signal segment into a pre-trained mutation signal recognition model to obtain an identification result indicating whether the signal segment is a mutation signal.
[0011] In one embodiment, the training process of the mutation signal recognition model includes:
[0012] Obtaining historical original signals;
[0013] Segmenting the historical original signals at the preset time interval to obtain a plurality of historical signal segments;
[0014] Extracting time-frequency features from each historical signal segment;
[0015] Use the between-class variance ratio method to determine the scores of each time-frequency feature of each historical signal segment respectively;
[0016] Select the time-frequency features with scores higher than the preset score from each time-frequency feature of each historical signal segment, and use the selected time-frequency features as the high-score features of the historical signal segment;
[0017] Perform tagging processing on the high-score features of each historical signal segment. The tag of each high-score feature is used to reflect whether the historical signal segment to which the high-score feature belongs is a burst signal;
[0018] Construct a training set and a validation set according to the tagged high-score features;
[0019] Use the training set for model training to obtain a converged mutation signal recognition model, and use the validation set to evaluate the performance of the converged mutation signal recognition model.
[0020] In one embodiment, before segmenting the historical original signal according to the preset time interval, the method further includes:
[0021] Perform augmentation processing on the historical original signal by adding Gaussian white noise to the historical original signal to obtain an augmented historical original signal;
[0022] Correspondingly, segmenting the historical original signal according to the preset time interval includes: segmenting the augmented historical original signal according to the preset time interval.
[0023] In one embodiment, before extracting time-frequency features from each historical signal segment, the method further includes:
[0024] Perform trend term interference removal processing on each historical signal segment;
[0025] Correspondingly, extracting time-frequency features from each historical signal segment includes: extracting time-frequency features from each historical signal segment after interference removal processing.
[0026] In one embodiment, the method further includes:
[0027] When the recognition result is that the signal segment is a mutation signal, store the signal segment and perform alarm processing.
[0028] In one embodiment, the mutation signal is an explosion signal, an industrial fault signal or a seismic monitoring signal.
[0029] In one embodiment, the mutation signal recognition model is an attention-enhanced channel propagation and aggregation delay neural network model.
[0030] According to a second aspect, the mutation signal recognition device based on deep learning provided by an embodiment of the present invention includes:
[0031] A signal acquisition module, configured to acquire an original signal;
[0032] A signal segmentation module, configured to segment the original signal at a preset time interval to obtain a plurality of signal segments;
[0033] A feature extraction module, configured to extract time-frequency features from each signal segment;
[0034] A feature scoring module, configured to determine the score of each time-frequency feature of each signal segment by using the between-class variance ratio method; wherein, the score of each time-frequency feature is used to characterize the feature discrimination degree corresponding to the time-frequency feature;
[0035] A feature selection module, configured to select time-frequency features with scores higher than a preset score from the time-frequency features of each signal segment, and use the selected time-frequency features as the high-score features of the signal segment;
[0036] A mutation recognition module, configured to input the high-score features of each signal segment into a pre-trained mutation signal recognition model to obtain a recognition result indicating whether the signal segment is a mutation signal.
[0037] In one embodiment, the device further includes:
[0038] A model training module, configured to perform the following steps:
[0039] Obtain historical original signals;
[0040] Segment the historical original signals at the preset time interval to obtain a plurality of historical signal segments;
[0041] Extract time-frequency features from each historical signal segment;
[0042] Determine the score of each time-frequency feature of each historical signal segment by using the between-class variance ratio method;
[0043] Select time-frequency features with scores higher than the preset score from the time-frequency features of each historical signal segment, and use the selected time-frequency features as the high-score features of the historical signal segment;
[0044] Perform tagging processing on the high-score features of each historical signal segment, and the tag of each high-score feature is used to indicate whether the historical signal segment to which the high-score feature belongs is a burst signal;
[0045] Construct a training set and a validation set according to the tagged high-score features;
[0046] Use the training set to train the model to obtain a converged mutation signal recognition model, and use the validation set to evaluate the performance of the converged mutation signal recognition model.
[0047] In one embodiment, before the model training module segments the historical original signal at the preset time interval, it is further configured to: perform augmentation processing on the historical original signal by adding Gaussian white noise to the historical original signal to obtain an augmented historical original signal; correspondingly, the process of segmenting the historical original signal at the preset time interval in the model training module includes: segmenting the augmented historical original signal at the preset time interval.
[0048] In one embodiment, before the model training module extracts time-frequency features from each historical signal segment, it is further configured to: perform trend term interference removal processing on each historical signal segment; correspondingly, the process of extracting time-frequency features from each historical signal segment in the model training module includes: extracting time-frequency features from each historical signal segment after the interference removal processing.
[0049] In one embodiment, the model training module further includes:
[0050] A signal processing module, configured to store the signal segment and perform an alarm process when the recognition result is that the signal segment is a mutation signal.
[0051] In one embodiment, the mutation signal is an explosion signal, an industrial fault signal, or a seismic monitoring signal.
[0052] In one embodiment, the mutation signal recognition model is an enhanced channel attention - propagation and aggregation delay neural network model.
[0053] According to a third aspect, an embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed on a computer, the computer is made to execute the method provided in the first aspect.
[0054] According to a fourth aspect, an embodiment of the present invention provides a computing device, including a memory and a processor. An executable code is stored in the memory. When the processor executes the executable code, the method provided in the first aspect is implemented.
[0055] The method, device, medium, and equipment for identifying mutation signals based on deep learning provided by the embodiments of the present invention first collect the original signal, then segment the original signal at a preset time interval, extract time-frequency features from each signal segment obtained by the segmentation process, score the time-frequency features, and then select high-score features. Finally, the high-score features of each signal segment are input into the mutation signal recognition model to obtain the recognition result indicating whether the signal segment is a mutation signal. First, in the above process, segmenting the signal can ensure a quick response to mutation signals and meet the real-time requirements in practical applications. Second, when selecting high-score features, it is achieved based on the between-class variance ratio method, and the between-class variance ratio method can effectively enhance the gap between target features and interference features, improve the sensitivity and discrimination of the model for target features, and thus improve the overall recognition accuracy. Additionally, using the mutation signal recognition model based on deep learning for mutation signal recognition can accurately identify mutation signals. It can be seen that the embodiments of the present invention have significant advantages in terms of recognition accuracy, real-time performance, etc., and can meet various requirements in practical applications. Description of the Drawings
[0056] Figure 1 It is a schematic flowchart of the method for identifying mutation signals based on deep learning in an embodiment of the present invention;
[0057] Figure 2 It is a structural block diagram of the device for identifying mutation signals based on deep learning in an embodiment of the present invention. Detailed Embodiments
[0058] In a first aspect, the embodiments of the present invention provide a method for identifying mutation signals based on deep learning. Refer to Figure 1 , the method includes the following steps S110 to S160:
[0059] S110. Collect the original signal;
[0060] Among them, the original signal can be a multi-channel signal input in real time.
[0061] S120. Segment the original signal at a preset time interval to obtain a plurality of signal segments;
[0062] It can be understood that segmenting the original signal to obtain a plurality of signal segments, and the time period corresponding to each signal segment is a time frame, that is, one signal segment is one signal frame. The segmentation process ensures the manageability and operability of the signal data.
[0063] S130. Extract time-frequency features from each signal segment;
[0064] It is understandable that S130 is to extract useful time-frequency features from the signal segment. Time-frequency analysis can display the features of the signal in terms of time and frequency, and can provide rich input data for the deep learning model.
[0065] Specifically, common time-frequency feature extraction methods can include short-time Fourier transform, Mel spectrogram, wavelet transform, etc.
[0066] S140. Use the between-class variance ratio method to determine the score of each time-frequency feature of each signal segment; wherein, the score of each time-frequency feature is used to characterize the feature discrimination degree corresponding to the time-frequency feature;
[0067] That is, use the between-class variance ratio method to score each extracted time-frequency feature and evaluate the discrimination degree of the time-frequency feature. The higher the score, the more useful the time-frequency feature is in distinguishing between mutant signals and non-mutant signals.
[0068] S150. Select the time-frequency features with scores higher than the preset score from each time-frequency feature of each signal segment, and use the selected time-frequency features as the high-score features of the signal segment;
[0069] That is, for each time-frequency feature of each signal segment, extract the high-score features from it. Of course, these high-score features can be adjusted and optimized to improve the effectiveness and representation ability of the high-score features.
[0070] S160. Input the high-score features of each signal segment into the pre-trained mutant signal recognition model to obtain the recognition result indicating whether the signal segment is a mutant signal.
[0071] Among them, after inputting the high-score features of a signal segment into the mutant signal recognition model, the model can output the recognition result, and the recognition result can reflect whether the signal segment is a mutant signal, so as to know whether a signal segment is a mutant signal.
[0072] In one embodiment, the training process of the mutant signal recognition model may include the following steps 1 to 8:
[0073] 1. Obtain historical original signals;
[0074] For example, the original signals in the past month. Specifically, the original signals can be obtained from multi-channel sensors or acquisition devices to ensure the integrity of the signals.
[0075] 2. Segment the historical original signals according to the preset time interval to obtain a plurality of historical signal segments;
[0076] 3. Extract time-frequency features from each historical signal segment;
[0077] 4. Use the between-class variance ratio method to determine the scores of each time-frequency feature of each historical signal segment;
[0078] 5. Select the time-frequency features with scores higher than the preset score from each time-frequency feature of each historical signal segment, and use the selected time-frequency features as the high-score features of the historical signal segment;
[0079] 6. Perform tagging on the high-score features of each historical signal segment. The tag of each high-score feature is used to indicate whether the historical signal segment to which the high-score feature belongs is a burst signal;
[0080] For example, if a historical signal segment is a mutation signal, the tags of the high-score features of the historical signal segment are all burst signals.
[0081] 7. Construct a training set and a validation set according to the tagged high-score features;
[0082] It can be understood that dividing the high-score features after feature extraction and selection into a training set and a validation set according to a certain proportion helps to ensure the effectiveness and generalization ability of model training.
[0083] 8. Use the training set to train the model to obtain a converged mutation signal recognition model, and use the validation set to evaluate the performance of the converged mutation signal recognition model.
[0084] It can be seen that the optimized high-score features are used to train the deep learning model for subsequent signal classification.
[0085] In one embodiment, the mutation signal recognition model can be an Emphasized Channel Attention Propagation and Aggregation Time Delay Neural Network model. The full English name of the Emphasized Channel Attention Propagation and Aggregation Time Delay Neural Network model is Emphasized Channel Attention Propagation and Aggregation Time Delay Neural Network, and the English abbreviation is ECAPA-TDNN. This model combines the SE-Res2Block structure, enhancing the feature representation ability and multi-scale feature processing ability.
[0086] In one embodiment, before segmenting the historical original signal at the preset time interval, the method may further include:
[0087] Perform augmentation processing on the historical original signal by adding Gaussian white noise to the historical original signal to obtain an augmented historical original signal;
[0088] Correspondingly, the step of segmenting the historical original signal according to the preset time interval includes: segmenting the augmented historical original signal according to the preset time interval.
[0089] It can be seen that data augmentation processing by adding Gaussian white noise to the collected historical original signal can increase the sample size.
[0090] In one embodiment, before extracting the time-frequency features from each historical signal segment, the method may further include: performing detrending interference removal processing on each historical signal segment;
[0091] Correspondingly, the step of extracting time-frequency features from each historical signal segment includes: extracting time-frequency features from each historical signal segment after detrending interference removal processing.
[0092] It can be seen that performing detrending interference removal processing on each historical signal segment can remove the detrending interference caused by explosion or other reasons.
[0093] It is understandable that operations such as amplification processing and interference removal are all preprocessing operations, which can provide clean and standardized data input for subsequent feature extraction and classification.
[0094] In one embodiment, the method further includes: when the recognition result indicates that the signal segment is a mutation signal, storing the signal segment and performing an alarm process.
[0095] It can be seen that storing and alarming mutation signals can promptly notify relevant personnel and ensure data protection and security.
[0096] In an actual scenario, post-processing can also be performed. Specifically, recording the original signal collected in S110, the recognition result obtained after the above S110 - S160, and the true verification result of whether the original signal is a burst signal can be used as samples for subsequent incremental training of the model, thereby improving the reliability and accuracy of the model.
[0097] In one embodiment, the mutation signal may be an explosion signal, an industrial fault signal, or a seismic monitoring signal.
[0098] It can be seen that the embodiments of the present invention can adapt to multi-channel signals of different types and sources, and are widely applied to mutation signal recognition tasks in various complex environments, such as explosion tests, seismic monitoring, industrial fault detection, etc.
[0099] For example, in an explosion test, a large amount of data is collected and mixed with a lot of useless data. To protect the data from the impact of the explosion, it is necessary to identify and store the data generated by the explosion first during the test. In the embodiments of the present invention, a feature extraction method is combined with a deep learning model to achieve the recognition of explosion signals, so as to improve the accuracy of explosion signal recognition.
[0100] Among them, the inter-class variance ratio (i.e., Inter-Class Variance Ratio) measures the separability of signals by calculating the variance ratio between different classes. By processing the features with the inter-class variance ratio, the gap between the target features and the interference features can be enlarged, thereby improving the score of the target features in model training and making the feature discrimination more significant. The inter-class variance ratio calculates the variance ratio between different classes, enhances the discrimination of features, and makes feature extraction more accurate.
[0101] Among them, the ECAPA-TDNN (i.e., Emphasized Channel Attention, Propagation and Aggregation Time Delay Neural Network) model has good performance in processing time series data and signal recognition. The SE-Res2Block structure in the model combines the residual structure with the squeeze-and-excitation module, enhances the expressiveness of features by adding residual connections between frame-level layers, and constructs hierarchical residual connections to process multi-scale features. That is, by introducing the SE-Res2Block structure, the ECAPA-TDNN model enhances the feature representation ability and multi-scale feature processing ability, making the model have higher recognition accuracy and real-time performance when processing complex signals. The TDNN structure can better capture the temporal information in the input features and has excellent performance under the task conditions of determining the explosion moment.
[0102] The embodiments of the present invention have the following beneficial effects:
[0103] (1) Improved recognition accuracy: Using deep learning algorithms and time-frequency features improved by the inter-class variance ratio, abnormal events in complex signals can be recognized more accurately. The ECAPA-TDNN model combined with the SE-Res2Block structure enhances the feature representation ability and multi-scale feature processing ability, making the model have higher recognition accuracy when processing complex signals. The method of inter-class variance ratio can effectively enhance the gap between target features and interference features, improve the sensitivity and discrimination of the model to target features, and thus improve the overall recognition accuracy.
[0104] (2) Real-time enhancement: By means of real-time segmentation and processing, it ensures a quick response to mutation signals and meets the requirements in practical applications. This method can process and identify mutation signals in real time during the explosion test, store data in time and give an alarm, effectively protecting the test data from the impact of the explosion. It can maintain stable real-time performance under high data volume conditions, ensuring a rapid response to mutation signals at critical moments.
[0105] (3) Adaptability and scalability: The model can adaptively adjust and learn newly collected samples, with good scalability and adaptability.
[0106] (4) By introducing deep learning algorithms, it can be continuously updated and optimized to improve the recognition ability for new and complex mutation signals.
[0107] (5) Multi-channel signal processing ability: The embodiment of the present invention can work effectively in a multi-channel signal acquisition system, distinguish and process mutation signals from different signal sources. Through the combination of feature extraction and deep learning models, it can effectively extract features and classify among multi-channel signals, reducing interference between signals. The multi-channel signal processing ability enables the embodiment of the present invention to perform excellently in complex environments and can simultaneously monitor and identify mutation signals from multiple signal sources.
[0108] (6) Robustness and reliability: The embodiment of the present invention improves robustness and reliability through the combination of deep learning algorithms and feature extraction methods. In the face of high noise and complex backgrounds, it can still maintain a high recognition accuracy and stability. Considering various situations in practical applications, it has good anti-interference ability and data processing ability, ensuring long-term stable operation.
[0109] It can be seen that the embodiment of the present invention has significant advantages in terms of recognition accuracy, real-time performance, adaptability, multi-channel signal processing ability, as well as the robustness and reliability of the system, and can meet various requirements in practical applications.
[0110] In summary, the embodiment of the present invention provides a method for identifying mutation signals with high accuracy, high real-time performance and strong adaptability by combining the high-score feature extraction method of the between-class variance ratio and the ECAPA-TDNN deep learning model. It can not only significantly improve the recognition accuracy and real-time response ability of burst signals, but also effectively handle the interference between multi-channel signals and adapt to the diversified mutation signal recognition tasks in complex environments. That is to say, the embodiment of the present invention can significantly improve the accuracy and real-time performance of burst signal recognition, can effectively distinguish mutation signals from non-burst signals, and give an alarm in time when a burst signal is recognized. Especially in a multi-channel signal acquisition system, it can improve the overall performance and adapt to the diversified mutation signals in complex environments.
[0111] Second aspect, an embodiment of the present invention provides a mutation signal recognition device based on deep learning. Refer to Figure 2 , the device 100 includes:
[0112] A signal acquisition module 110, configured to acquire an original signal;
[0113] A signal segmentation module 120, configured to segment the original signal at a preset time interval to obtain a plurality of signal segments;
[0114] A feature extraction module 130, configured to extract time-frequency features from each signal segment;
[0115] A feature scoring module 140, configured to determine the score of each time-frequency feature of each signal segment by using the between-class variance ratio method; wherein, the score of each time-frequency feature is used to characterize the feature discrimination degree corresponding to the time-frequency feature;
[0116] A feature selection module 150, configured to select time-frequency features with scores higher than a preset score from the time-frequency features of each signal segment, and use the selected time-frequency features as the high-score features of the signal segment;
[0117] A mutation recognition module 160, configured to input the high-score features of each signal segment into a pre-trained mutation signal recognition model to obtain a recognition result indicating whether the signal segment is a mutation signal.
[0118] In one embodiment, the device further includes:
[0119] A model training module, configured to perform the following steps:
[0120] Obtain historical original signals;
[0121] Segment the historical original signals at the preset time interval to obtain a plurality of historical signal segments;
[0122] Extract time-frequency features from each historical signal segment;
[0123] Use the between-class variance ratio method to determine the score of each time-frequency feature of each historical signal segment;
[0124] Select time-frequency features with scores higher than the preset score from the time-frequency features of each historical signal segment, and use the selected time-frequency features as the high-score features of the historical signal segment;
[0125] Perform tagging processing on the high-score features of each historical signal segment, and the tag of each high-score feature is used to indicate whether the historical signal segment to which the high-score feature belongs is a burst signal;
[0126] Construct a training set and a validation set according to each high-score feature with labels;
[0127] Use the training set to train the model to obtain a converged mutation signal recognition model, and use the validation set to evaluate the performance of the converged mutation signal recognition model.
[0128] In one embodiment, before the model training module segments the historical original signal at the preset time interval, it is further used for: performing augmentation processing on the historical original signal by adding Gaussian white noise to the historical original signal to obtain an augmented historical original signal; correspondingly, the process of segmenting the historical original signal at the preset time interval in the model training module includes: segmenting the augmented historical original signal at the preset time interval.
[0129] In one embodiment, before the model training module extracts time-frequency features from each historical signal segment, it is further used for: performing trend term interference removal processing on each historical signal segment; correspondingly, the process of extracting time-frequency features from each historical signal segment in the model training module includes: extracting time-frequency features from each historical signal segment after interference removal processing.
[0130] In one embodiment, the model training module further includes:
[0131] A signal processing module, configured to store the signal segment and perform an alarm process when the recognition result is that the signal segment is a mutation signal.
[0132] In one embodiment, the mutation signal is an explosion signal, an industrial fault signal, or a seismic monitoring signal.
[0133] In one embodiment, the mutation signal recognition model is an enhanced channel attention - propagation and aggregation delay neural network model.
[0134] It can be understood that the explanations, specific implementation manners, beneficial effects, examples, etc. of the relevant content in the device provided in the embodiments of the present invention can refer to the corresponding parts in the method provided in the first aspect, and will not be elaborated here.
[0135] In a third aspect, an embodiment of the present invention provides a computer-readable medium, on which computer instructions are stored, and when the computer instructions are executed by a processor, the processor is caused to execute the method provided in the first aspect.
[0136] Specifically, a system or device equipped with a storage medium can be provided. On this storage medium, software program code for implementing the functions of any one of the above embodiments is stored, and the computer (or CPU or MPU) of the system or device is made to read and execute the program code stored in the storage medium.
[0137] In this case, the program code read from the storage medium itself can implement the functions of any one of the above embodiments. Therefore, the program code and the storage medium storing the program code constitute a part of the present invention.
[0138] Examples of the storage medium for providing the program code include floppy disks, hard disks, magneto-optical disks, optical disks (such as CD-ROM, CD-R, CD-RW, DVD-ROM, DVD-RAM, DVD-RW, DVD+RW), magnetic tapes, non-volatile memory cards, and ROMs. Optionally, the program code can be downloaded from a server computer via a communication network.
[0139] Furthermore, it should be clear that not only can the functions of any one of the above embodiments be realized by executing the program code read by the computer, but also by causing an operating system or the like operating on the computer based on the instructions of the program code to complete part or all of the actual operations.
[0140] In addition, it can be understood that the program code read from the storage medium is written into the memory provided in an expansion board inserted into the computer or into the memory provided in an expansion module connected to the computer, and then based on the instructions of the program code, the CPU or the like installed on the expansion board or the expansion module is made to execute part and all of the actual operations, thereby realizing the functions of any one of the above embodiments.
[0141] It can be understood that the explanations, specific implementation manners, beneficial effects, examples, etc. of the content related to the computer-readable medium provided by the embodiments of the present invention can be referred to the corresponding parts in the method provided in the first aspect, and will not be elaborated here.
[0142] In a fourth aspect, an embodiment of this specification provides a computing device, including a memory and a processor. An executable code is stored in the memory. When the processor executes the executable code, the method in any one of the embodiments in the specification is implemented.
[0143] It can be understood that the explanations, specific implementation manners, beneficial effects, examples, etc. of the content related to the computing device provided by the embodiments of the present invention can be referred to the corresponding parts in the method provided in the first aspect, and will not be elaborated here.
[0144] Each embodiment in this specification is described in a progressive manner. For the same or similar parts among the embodiments, reference can be made to each other. Each embodiment focuses on the differences from other embodiments. In particular, for the device embodiments, since they are basically similar to the method embodiments, the description is relatively simple. For the relevant parts, reference can be made to the partial description of the method embodiments.
[0145] Those skilled in the art should be able to realize that in one or more of the above examples, the functions described in the present invention can be implemented by hardware, software, add-ons, or any combination thereof. When implemented using software, these functions can be stored in a computer-readable medium or transmitted as one or more instructions or codes on a computer-readable medium.
[0146] The specific implementation manners described above further elaborate on the purpose, technical solutions, and beneficial effects of the present invention. It should be understood that the above is only the specific implementation manner of the present invention and is not used to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made on the basis of the technical solutions of the present invention shall be included in the protection scope of the present invention.
Claims
1. A method for identifying mutation signals based on deep learning, characterized in that, Including: Collecting the original signal; Segmenting the original signal at a preset time interval to obtain a plurality of signal segments; Extracting time-frequency features from each signal segment; Using the between-class variance ratio method to determine the score of each time-frequency feature of each signal segment; wherein, the score of each time-frequency feature is used to characterize the feature discrimination degree corresponding to the time-frequency feature; Selecting the time-frequency features with scores higher than the preset score from the time-frequency features of each signal segment, and taking the selected time-frequency features as the high-score features of the signal segment; Inputting the high-score features of each signal segment into a pre-trained mutation signal recognition model to obtain an identification result indicating whether the signal segment is a mutation signal.
2. The method according to claim 1, characterized in that, The training process of the mutation signal recognition model includes: Obtaining historical original signals; Segmenting the historical original signals at the preset time interval to obtain a plurality of historical signal segments; Extracting time-frequency features from each historical signal segment; Using the between-class variance ratio method to determine the score of each time-frequency feature of each historical signal segment; Selecting the time-frequency features with scores higher than the preset score from the time-frequency features of each historical signal segment, and taking the selected time-frequency features as the high-score features of the historical signal segment; Labeling the high-score features of each historical signal segment, and the label of each high-score feature is used to indicate whether the historical signal segment to which the high-score feature belongs is a burst signal; Constructing a training set and a validation set according to the labeled high-score features; Using the training set to perform model training to obtain a converged mutation signal recognition model, and using the validation set to evaluate the performance of the converged mutation signal recognition model.
3. The method according to claim 2, wherein Before segmenting the historical original signals at the preset time interval, the method further includes: Performing augmentation processing on the historical original signals by adding Gaussian white noise to the historical original signals to obtain augmented historical original signals; Correspondingly, segmenting the historical original signals at the preset time interval includes: segmenting the augmented historical original signals at the preset time interval.
4. The method according to claim 2, wherein Before extracting time-frequency features from each historical signal segment, the method further includes: Performing trend term interference removal processing on each historical signal segment; Correspondingly, extracting time-frequency features from each historical signal segment includes: extracting time-frequency features from each historical signal segment after interference removal processing.
5. The method according to claim 1, wherein Also including: When the recognition result is that the signal segment is a mutation signal, storing the signal segment and performing alarm processing.
6. The method according to claim 1, wherein The mutation signal is an explosion signal, an industrial fault signal or a seismic monitoring signal.
7. The method according to claim 2, wherein The mutation signal recognition model is an enhanced channel attention - propagation and aggregation delay neural network model.
8. A mutation signal recognition device based on deep learning, characterized in that, Including: A signal acquisition module for collecting the original signal; A signal segmentation module for segmenting the original signal at a preset time interval to obtain a plurality of signal segments; A feature extraction module for extracting time-frequency features from each signal segment; A feature scoring module, which is used to determine the respective scores of each time-frequency feature of each signal segment by using the between-class variance ratio method; wherein, the score of each time-frequency feature is used to characterize the feature discrimination degree corresponding to the time-frequency feature. A feature selection module, which is used to select time-frequency features with scores higher than a preset score from each time-frequency feature of each signal segment, and use the selected time-frequency features as the high-score features of the signal segment. A mutation recognition module, which is used to input the high-score features of each signal segment into a pre-trained mutation signal recognition model to obtain an identification result indicating whether the signal segment is a mutation signal.
9. A computer-readable storage medium, characterized in that, It stores a computer program, and when the computer program is executed on a computer, it causes the computer to execute the method described in any one of claims 1 to 7.
10. A computing device, characterized in that, It includes a memory and a processor. An executable code is stored in the memory, and when the processor executes the executable code, it implements the method described in any one of claims 1 to 7.