Deep learning seismic event classification method and system based on multi-feature fusion
Through the deep learning method of multi-feature fusion, combined with prior knowledge and data enhancement technology, the classification accuracy of earthquake events is improved, especially the recognition ability of non-natural earthquakes, and the problems of low accuracy and manpower in the existing technology are solved, and the problems of low accuracy and labor-intensiveness are achieved. Efficient and intelligent earthquake event classification.
Patent Information
- Application Number
- CN202410982427.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-22
- Publication Date
- 2024-11-15
AI Technical Summary
The prior art has low accuracy in the classification of earthquake events, especially the classification accuracy of non-natural earthquake events, and requires a lot of manpower to process waveforms, which limits large-scale promotion and use.
Using a deep learning method based on multi-feature fusion, seismic waveform data is collected through multiple stations, preprocessed and inputted to pre-trained classification network for classification, including feature calculation based on prior knowledge, feature extraction of XGBoost, and Bi-LSTM waveform feature extraction, data augmentation generated by rare categories, and feature selection networks of Attention, combined with voting mechanisms to determine event types.
The classification accuracy of natural earthquakes and non-natural earthquake events has been improved, especially the identification accuracy of non-natural earthquakes has been balanced through data augmentation, and efficient and intelligent seismic event classification has been achieved.
Smart Images

Figure CN118965100A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of seismic data processing, and specifically relates to a deep learning seismic event classification method and system based on multi-feature fusion. Background Art
[0002] Earthquake is a natural phenomenon caused by plate movement. Its focal mechanism is the release of a large amount of energy after the collision of plates. The most destructive natural disaster is earthquake, which occurs instantaneously and causes a large number of casualties. At present, the seismic network center detects the occurrence of earthquakes based on the seismic waves received by the stations, and phenomena caused by non-geological structures such as explosions, blasting, collapse and volcanoes will also produce waveform data similar to seismic waves. With the rapid development of China's digital seismic observation system technology, the density of stations has gradually increased, and the earthquake monitoring capability has been significantly improved. In addition to being able to monitor various types of tectonic earthquake events, it can also effectively capture more signals of various types of non-tectonic events. Classifying tectonic events and non-tectonic events can provide accurate nature information in the first time after the earthquake, which is of great significance to reducing earthquake losses, reducing earthquake disasters and reducing casualties.
[0003] The difference in features between seismic waveforms and non-tectonic event waveforms is the key factor in distinguishing earthquakes from non-tectonic events such as explosions and collapses. With the increase in the number of stations in the network, the collection of seismic waveform data has become more complete, and the preprocessing of seismic waveform data has become more important. Only by using reasonable preprocessing methods for seismic waveform data can the ability to classify events be improved. At present, there are some methods for processing seismic waveform data. Using event waveforms to quickly classify and judge seismic events is of great significance to subsequent earthquake rapid reporting.
[0004] Manual analysis of earthquake waveform features consumes a lot of manpower and material resources in the process of earthquake event classification. At present, there are a large number of labeled tectonic earthquake events and non-tectonic events. The recently emerging deep learning technology that uses a large amount of data to extract features for learning can be applied to the problem of earthquake event classification to improve the efficiency of earthquake event classification and save resources.
[0005] With the rapid development of deep learning technology, deep learning methods have been widely used in the field of seismology. Although the current methods can roughly classify earthquake events, the accuracy is not high. Due to the small number of non-natural earthquake events, the current model has low classification accuracy for non-natural earthquake events. Even if there are some more accurate methods, it takes a lot of manpower to process the waveforms of earthquake events, which is not conducive to large-scale promotion and use. Summary of the invention
[0006] In view of the shortcomings of the existing technology, this application proposes a deep learning earthquake event classification method and system based on multi-feature fusion.
[0007] In the first aspect, the present application proposes a deep learning earthquake event classification method based on multi-feature fusion, including:
[0008] Seismic waveform data are collected through multiple stations;
[0009] Preprocessing the seismic waveform data;
[0010] Input the preprocessed seismic waveform data into a pretrained classification network for classification to obtain a seismic event classification result; the pretrained classification network includes: a feature calculation network based on prior knowledge, a feature extraction network based on XGBoost, a waveform feature extraction network based on Bi-LSTM, a data enhancement network based on rare category generation, a feature selection network based on Attention, and a classification network;
[0011] The earthquake event classification results of all stations are voted, the earthquake event type is determined according to the voting results, and the earthquake event type is output and displayed.
[0012] The feature calculation network based on prior knowledge includes: using the concept of seismic waveform features proposed by experts to perform feature calculation on pre-processed seismic waveform data to obtain a first feature, which reflects the multi-angle characteristics of the signal in the time-frequency domain and energy.
[0013] The XGBoost-based feature extraction network includes: taking the first feature as input, using a pre-trained XGBoost model to evaluate each feature contained in the first feature, obtaining the weight of each feature in the event classification process, and taking the weighted first feature as output.
[0014] The Bi-LSTM-based waveform feature extraction network includes: taking preprocessed seismic waveform data as input, using a pre-trained Bi-LSTM model to extract a second feature, wherein the second feature is a time domain feature, and taking the second feature as output.
[0015] The data enhancement network generated based on rare categories includes:
[0016] The first feature with weight is combined with the second feature to obtain the third feature;
[0017] The third feature is input into a pre-trained data enhancement network to obtain a fourth feature.
[0018] The Attention-based feature selection network includes: inputting the fourth feature into a pre-trained Attention model to obtain a fifth feature.
[0019] The classification network includes: inputting the fifth feature into a pre-trained fully connected layer to obtain an earthquake event classification result.
[0020] In the second aspect, the present application proposes a deep learning earthquake event classification system based on multi-feature fusion, including:
[0021] A data acquisition module, used for collecting seismic waveform data through multiple stations;
[0022] A preprocessing module, used for preprocessing the seismic waveform data;
[0023] An earthquake event classification module is used to input the preprocessed earthquake waveform data into a pretrained classification network for classification to obtain earthquake event classification results; the pretrained classification network includes: a feature calculation network based on prior knowledge, a feature extraction network based on XGBoost, a waveform feature extraction network based on Bi-LSTM, a data enhancement network based on rare category generation, a feature selection network based on Attention, and a classification network;
[0024] The visualization output module is used to vote on the earthquake event classification results of all stations, determine the earthquake event type according to the voting results, and output and display the earthquake event type.
[0025] In a third aspect, the present application proposes an electronic device comprising: one or more processors, and a memory, wherein the memory is used to store instructions, and when the instructions are executed by the one or more processors, the one or more processors execute the deep learning earthquake event classification method based on multi-feature fusion.
[0026] In a fourth aspect, the present application proposes a computer-readable storage medium storing executable instructions, which, when executed, enable a processor to execute the deep learning earthquake event classification method based on multi-feature fusion.
[0027] Beneficial effects:
[0028] The present application proposes a deep learning earthquake event classification method and system based on multi-feature fusion. The present application improves the accuracy of natural earthquake and non-natural earthquake event classification by fusing physical features and data enhancement. The feature calculation module is implemented using prior knowledge, and the weight information of each feature is obtained according to XGBoost for the earthquake event classification process. The number of non-natural earthquake events is less than that of natural earthquakes. During training, it is easy to identify non-natural earthquakes as natural earthquakes. In order to balance the number of the two types of earthquakes, the present application uses feature displacement to perform data enhancement on non-natural earthquakes, thereby improving the recognition accuracy of non-natural earthquakes. BRIEF DESCRIPTION OF THE DRAWINGS
[0029] Figure 1 A flow chart of a deep learning earthquake event classification method based on multi-feature fusion in an embodiment of the present application;
[0030] Figure 2 A model structure diagram of a deep learning earthquake event classification method based on multi-feature fusion in an embodiment of the present application;
[0031] Figure 3 A schematic diagram of a flow chart of a deep learning earthquake event classification method based on multi-feature fusion according to an embodiment of the present application;
[0032] Figure 4 A principle block diagram of a deep learning earthquake event classification device based on multi-feature fusion in an embodiment of the present application. DETAILED DESCRIPTION
[0033] The specific implementation of the present application is further described in detail below in conjunction with the drawings and examples.
[0034] Although the current earthquake event method can roughly classify earthquake events, the accuracy is not high. Due to the small number of non-natural earthquake events, the current model has low classification accuracy for non-natural earthquake events. This application proposes a deep learning earthquake event classification method and system based on multi-feature fusion. Under the premise that the number of non-natural earthquake events is small, this method can classify natural earthquakes and non-natural earthquakes, where non-natural earthquakes include explosions, collapses, and volcanoes. The model uses prior knowledge, integrates physical features with seismic waveform features, and uses data enhancement to expand the number of small sample non-natural earthquake data, thereby realizing an intelligent earthquake event classification system and improving the accuracy and intelligence of earthquake event classification.
[0035] Embodiment 1:
[0036] This embodiment proposes a deep learning earthquake event classification method based on multi-feature fusion, such as Figure 1 to Figure 3 As shown, including:
[0037] Step S1: collecting seismic waveform data through multiple stations;
[0038] In this embodiment, seismic waveform data is collected through multiple stations, and earthquake catalog information is queried to determine the accurate arrival time of the three-component seismic P wave. After determining the first arrival time of the P wave, the data from 100 seconds before to 2 minutes after the starting point of each component is cut as the seismic waveform data. The sampling frequency of the data is 100 Hz, with a total of 22,000 sampling points.
[0039] Step S2: preprocessing the seismic waveform data;
[0040] In this embodiment, the seismic waveform data is preprocessed. The station may be affected by various external factors during the process of recording the seismic waveform data, resulting in signal interruption. Therefore, it is necessary to remove the seismic waveform data containing interruptions and low signal-to-noise ratio. In addition, during the propagation of seismic waves, their waveforms may be affected by greater noise, resulting in unclear characteristics of the seismic waveform. Therefore, it is necessary to reduce noise and filter the waveform before use. The waveform data is divided by the sensitivity of the seismometer to correct the error caused by the instrument's own influencing factors, and then the waveform is subjected to unified data processing operations, including filtering, mean detrending and normalization operations.
[0041] Step S3: input the preprocessed seismic waveform data into a pre-trained classification network for classification to obtain a seismic event classification result; the pre-trained classification network includes: a feature calculation network based on prior knowledge, a feature extraction network based on XGBoost, a waveform feature extraction network based on Bi-LSTM, a data enhancement network based on rare category generation, a feature selection network based on Attention, and a classification network;
[0042] In this embodiment, it is first necessary to preprocess the historical seismic waveform data, input the preprocessed historical seismic waveform data into the pre-trained classification network, train various networks, and obtain a pre-trained XGBoost model, a pre-trained Bi-LSTM model, a pre-trained data enhancement network, and a pre-trained Attention model, thereby completing the pre-training of various networks, and finally obtaining: a feature calculation network based on prior knowledge, a feature extraction network based on XGBoost, a waveform feature extraction network based on Bi-LSTM, a data enhancement network based on rare category generation, a feature selection network based on Attention, and a classification network.
[0043] Among them, the XGBoost model (eXtreme Gradient Boosting) usually corrects the errors of the previous learners by continuously adding new weak learners to improve the accuracy of the model. This application uses a pre-trained XGBoost model to obtain the weights of the extracted features. This allows the features proposed by experts to be more accurately applied in earthquake event classification;
[0044] Bi-LSTM model (Bidirectional Long Short-Term Memory, bidirectional long short-term memory neural network) Bi-LSTM model is a neural network structure that combines forward and backward LSTM (Long Short-Term Memory, long short-term memory neural network). This model can simultaneously consider the front and back information of the input sequence, thereby providing richer contextual understanding when processing certain tasks. Bi-LSTM model is usually used in natural language processing. When processing natural language processing tasks, Bi-LSTM model can capture the dependency between words in a sentence. In this application, seismic waveform data is a data point feature sorted in time, so there is a mutual dependency between data points, and the front and back information of the waveform can be extracted by Bi-LSTM model, so Bi-LSTM model is used to extract waveform features, and waveform features are combined with features calculated by experts, and applied to earthquake event classification, and good technical effects are achieved, and the accuracy of earthquake event classification is improved.
[0045] Attention Model (Attention Model, attention mechanism model), this model makes extensive use of the self-attention mechanism to learn text representation. This model is widely used in the field of natural language processing (NLP). By using the self-attention mechanism, it can better capture the dependencies in the text. In the field of earthquake event classification, due to the temporal attribute characteristics of earthquake waveforms, the Attention mechanism can also capture important information about the changes in waveform data, which is very important for earthquake event classification. Therefore, this application uses the Attention model to extract the characteristics of the waveform, and has achieved good technical results, improving the accuracy of earthquake event classification.
[0046] The feature calculation network based on prior knowledge includes: using the concept of seismic waveform features proposed by experts to perform feature calculation on the preprocessed seismic waveform data to obtain a first feature, wherein the first feature reflects the multi-angle features of the signal in the time-frequency domain and energy. In this embodiment, the concept of seismic waveform features proposed by experts is used to perform feature calculation on the preprocessed seismic waveform data, reflecting the features of the signal in multiple angles such as the time-frequency domain and energy.
[0047] The XGBoost-based feature extraction network includes: taking the first feature as input, using a pre-trained XGBoost model to evaluate each feature contained in the first feature, obtaining the weight of each feature in the event classification process, and taking the weighted first feature as output.
[0048] In this embodiment, the preprocessed seismic waveform data is input into a feature calculation network based on prior knowledge, multiple eigenvalues of each seismic waveform are calculated, and then the multiple eigenvalues of each seismic waveform are input into a feature extraction network based on XGBoost, and the weight of the feature in the classification process is calculated, and the output is a multi-channel two-dimensional feature tensor; the multiple features extracted in this step provide the effect of increasing or reducing the accuracy when classifying earthquake events, and the weights they occupy are obtained and applied together with the eigenvalues to subsequent training.
[0049] The Bi-LSTM-based waveform feature extraction network includes: taking preprocessed seismic waveform data as input, using a pre-trained Bi-LSTM model to extract a second feature, wherein the second feature is a time domain feature, and taking the second feature as output.
[0050] In this embodiment, characteristic information of historical seismic waveforms is extracted in bidirectional time and the weights of the network are trained to obtain a deep learning network model for feature extraction based on Bi-LSTM; the preprocessed seismic waveform data to be classified is input into a waveform feature extraction network based on Bi-LSTM, and a second feature is output, which is a multi-channel two-dimensional feature tensor.
[0051] The above two feature extraction networks (i.e., the feature extraction network based on XGBoost and the waveform feature extraction network based on Bi-LSTM) process the preprocessed seismic waveform data based on prior knowledge, extract the corresponding features, and then splice them. They will enter the data enhancement stage, using feature displacement to generate more small sample data to balance the weight of non-natural earthquakes in the classification process. After that, the feature data after data enhancement is input into the feature selection network, and feature selection is performed using Attention. After feature selection, it is input into the classification model to achieve multi-classification tasks.
[0052] The data enhancement network generated based on rare categories includes:
[0053] The first feature with weight is combined with the second feature to obtain the third feature;
[0054] The third feature is input into a pre-trained data enhancement network to obtain a fourth feature.
[0055] In this embodiment, for the historical preprocessed seismic waveform data, the two multi-channel two-dimensional feature tensors (i.e., the first feature and the second feature) extracted are merged into a multi-feature two-dimensional feature tensor (i.e., the third feature), and input into a data enhancement network generated based on rare categories, and the network weights of the network are trained to obtain a data enhancement network generated based on rare categories; the multi-channel two-dimensional feature tensor extracted from the preprocessed seismic waveform data to be classified is input into the data enhancement network model generated based on rare categories to obtain the fourth feature, specifically a single-channel one-dimensional feature tensor. The pre-trained data enhancement network uses feature displacement to generate more sample data.
[0056] The Attention-based feature selection network includes: inputting the fourth feature into a pre-trained Attention model to obtain a fifth feature.
[0057] In this embodiment, the fourth feature extracted from the historical preprocessed seismic waveform data is input into the feature selection network based on Attention, the weight of the network is trained, and the deep learning network model for earthquake time classification based on Attention is obtained, that is, the feature selection network based on Attention. The fourth feature extracted from the preprocessed seismic waveform data to be classified is input into the feature selection network based on Attention, and a single-channel one-dimensional feature tensor, that is, the fifth feature, is output.
[0058] The classification network includes: inputting the fifth feature into a pre-trained fully connected layer to obtain an earthquake event classification result.
[0059] In this embodiment, the fifth feature extracted from the historical seismic waveform data is input into a fully connected layer, the weight of the fully connected layer is trained, and the parameters of the fully connected layer are obtained; the fifth feature extracted from the preprocessed seismic waveform data to be classified, that is, the one-dimensional data tensor of a single channel, is input into the pre-trained fully connected layer to obtain the earthquake event classification result.
[0060] Step S4: Vote for the earthquake event classification results of all stations, determine the earthquake event type according to the voting results, and output and display the earthquake event type.
[0061] In this embodiment, the earthquake event classification result can be obtained by the above method. As the most destructive natural disaster, earthquake can correctly classify earthquake events to instruct relevant departments to take different emergency measures according to different earthquake events to reduce the hazards of earthquakes. Compared with the prior art, the technical solution proposed in this application integrates physical features and the weights of physical features into the waveform features extracted by the network, splices the two parts of features and enhances the data, and can comprehensively train multiple factors when classifying earthquake events. The deep learning earthquake event classification method based on multi-feature fusion described in this application is adopted, and the earthquake waveform data recorded by seismic stations in China from January 2009 to December 2019 provided by the China Earthquake Networks Center are used for training and testing. From the experimental results, the earthquake event classification accuracy of the method of this application is 94%, and a good classification ability has been achieved in the direction of earthquake event classification. This application is simple to deploy, effective in detection, and meets the requirements of application. When the number of non-natural earthquake events is smaller than the number of natural earthquakes, it is easy to identify non-natural earthquakes as natural earthquakes during training. In order to balance the number of the two types of earthquakes, this application uses characteristic displacement to perform data enhancement on non-natural earthquakes, thereby improving the recognition accuracy of non-natural earthquakes.
[0062] Embodiment 2:
[0063] This embodiment proposes a deep learning earthquake event classification system based on multi-feature fusion, such as Figure 4 As shown, it includes: a data acquisition module, a preprocessing module, a seismic event classification module and a visualization output module, and the data acquisition module, the preprocessing module, the seismic event classification module and the visualization output module are connected in sequence;
[0064] A data acquisition module, used for collecting seismic waveform data through multiple stations;
[0065] A preprocessing module, used for preprocessing the seismic waveform data;
[0066] An earthquake event classification module is used to input the preprocessed earthquake waveform data into a pretrained classification network for classification to obtain earthquake event classification results; the pretrained classification network includes: a feature calculation network based on prior knowledge, a feature extraction network based on XGBoost, a waveform feature extraction network based on Bi-LSTM, a data enhancement network based on rare category generation, a feature selection network based on Attention, and a classification network;
[0067] The visualization output module is used to vote on the earthquake event classification results of all stations, determine the earthquake event type according to the voting results, and output and display the earthquake event type.
[0068] Embodiment 3:
[0069] This embodiment proposes an electronic device, including: one or more processors, and a memory, wherein the memory is used to store instructions, and when the instructions are executed by the one or more processors, the one or more processors execute the deep learning earthquake event classification method based on multi-feature fusion.
[0070] The electronic device may be a mobile phone, a computer or a tablet computer, etc., including a memory and a processor, wherein a computer program is stored on the memory, and when the computer program is executed by the processor, the deep learning earthquake event classification method based on multi-feature fusion as described in the embodiment is implemented. It is understood that the electronic device may also include an input / output (I / O) interface and a communication component.
[0071] The processor is used to execute all or part of the steps in the deep learning earthquake event classification method based on multi-feature fusion as described in the above embodiment. The memory is used to store various types of data, which may include instructions of any application or method in the electronic device, as well as data related to the application.
[0072] The processor can be an application specific integrated circuit (ASIC), a digital signal processor (DSP), a programmable logic device (PLD), a field programmable gate array (FPGA), a controller, a microcontroller, a microprocessor or other electronic components, and is used to execute the deep learning earthquake event classification method based on multi-feature fusion described in the above embodiments.
[0073] Embodiment 4:
[0074] This embodiment provides a computer-readable storage medium storing executable instructions. When the instructions are executed and implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium.
[0075] The computer software product is stored in a storage medium and includes a number of instructions for enabling a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the deep learning earthquake event classification method based on multi-feature fusion described in each embodiment of the present application.
[0076] The aforementioned storage media include: flash memory, hard disk, multimedia card, card-type memory (for example, SD (Secure Digital Memory Card) or DX (Memory Data Register, MDR abbreviation, memory data register) memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, CD, server, APP (Application, abbreviation of application software) application store and other media that can store program verification codes, on which computer programs are stored. When the computer program is executed by the processor, the various steps of the above-mentioned deep learning earthquake event classification method based on multi-feature fusion can be implemented.
[0077] The various embodiments in the present application are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.
[0078] The protection scope of the present application is not limited to the above-mentioned embodiments. Obviously, those skilled in the art can make various changes and modifications to the present disclosure without departing from the scope and spirit of the present disclosure. If these changes and modifications fall within the scope of the claims of the present disclosure and their equivalents, the intention of the present disclosure also includes these changes and modifications.
Claims
1. A deep learning earthquake event classification method based on multi-feature fusion, characterized in that: include: Seismic waveform data are collected through multiple stations; Preprocessing the seismic waveform data; The preprocessed seismic waveform data is input into the pre-trained classification network for classification to obtain the seismic event classification result; The pre-trained classification network includes: a feature calculation network based on prior knowledge, a feature extraction network based on XGBoost, a waveform feature extraction network based on Bi-LSTM, a data enhancement network based on rare category generation, a feature selection network based on Attention, and a classification network; The earthquake event classification results of all stations are voted, the earthquake event type is determined according to the voting results, and the earthquake event type is output and displayed.
2. The deep learning earthquake event classification method based on multi-feature fusion according to claim 1 is characterized in that: The feature calculation network based on prior knowledge includes: using the concept of seismic waveform features proposed by experts to perform feature calculation on pre-processed seismic waveform data to obtain a first feature, which reflects the multi-angle characteristics of the signal in the time-frequency domain and energy.
3. The deep learning earthquake event classification method based on multi-feature fusion according to claim 2 is characterized in that: The XGBoost-based feature extraction network includes: taking the first feature as input, using a pre-trained XGBoost model to evaluate each feature contained in the first feature, obtaining the weight of each feature in the event classification process, and taking the weighted first feature as output.
4. The deep learning earthquake event classification method based on multi-feature fusion according to claim 3 is characterized in that: The Bi-LSTM-based waveform feature extraction network includes: taking preprocessed seismic waveform data as input, using a pre-trained Bi-LSTM model to extract a second feature, wherein the second feature is a time domain feature, and taking the second feature as output.
5. The deep learning earthquake event classification method based on multi-feature fusion according to claim 4 is characterized in that: The data enhancement network generated based on rare categories includes: The first feature with weight is combined with the second feature to obtain the third feature; The third feature is input into a pre-trained data enhancement network to obtain a fourth feature.
6. The deep learning earthquake event classification method based on multi-feature fusion according to claim 5 is characterized in that: The Attention-based feature selection network includes: inputting the fourth feature into a pre-trained Attention model to obtain a fifth feature.
7. The deep learning earthquake event classification method based on multi-feature fusion according to claim 6 is characterized in that: The classification network includes: inputting the fifth feature into a pre-trained fully connected layer to obtain an earthquake event classification result.
8. A deep learning earthquake event classification system based on multi-feature fusion, characterized in that: include: A data acquisition module, used for collecting seismic waveform data through multiple stations; A preprocessing module, used for preprocessing the seismic waveform data; The earthquake event classification module is used to input the preprocessed earthquake waveform data into the pre-trained classification network for classification to obtain the earthquake event classification result; The pre-trained classification network includes: a feature calculation network based on prior knowledge, a feature extraction network based on XGBoost, a waveform feature extraction network based on Bi-LSTM, a data enhancement network based on rare category generation, a feature selection network based on Attention, and a classification network; The visualization output module is used to vote on the earthquake event classification results of all stations, determine the earthquake event type according to the voting results, and output and display the earthquake event type.
9. An electronic device, characterized in that: include: One or more processors, and a memory, wherein the memory is used to store instructions, and when the instructions are executed by the one or more processors, the one or more processors execute the deep learning seismic event classification method based on multi-feature fusion as described in any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that: It stores executable instructions, which, when executed, enable the processor to execute the deep learning earthquake event classification method based on multi-feature fusion as described in any one of claims 1 to 7.
