Seismic event detection method based on multi-scale transformation of seabed application scene
Through multi-scale transformation and multi-level collaborative design, combined with joint verification of seabed seismometers, buoy nodes and shore-based processing centers, the real-time and accuracy issues in seabed seismic event detection were resolved, enabling rapid and accurate seismic event detection.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHONGQING GEOLOGICAL INSTR FACTORY
- Filing Date
- 2025-12-31
- Publication Date
- 2026-04-10
AI Technical Summary
Existing technologies struggle to balance real-time performance and detection accuracy in submarine seismic event detection. They fail to fully utilize information from multiple sensors, do not integrate the complementary advantages of triaxial velocity, triaxial acceleration, and hydrophone signals, and suffer from poor communication compatibility, resulting in slow detection loop closure and difficulty in coping with strong time-varying noise and sensor performance drift.
A multi-scale transformation method is adopted, which collects multi-channel signals through seabed seismometers, performs multi-scale time-frequency transformation processing, constructs fused feature vectors, and conducts joint verification at buoy nodes and shore-based processing centers. A lightweight 1D-CNN model and Bayesian global confidence fusion are used to achieve multi-level collaborative design and deep integration of multi-source information.
It enables rapid response and accurate identification of earthquake events, improves detection speed and accuracy, adapts to the transmission requirements of underwater acoustic communication scenarios, reduces data redundancy, and improves the reliability and accuracy of detection.
Smart Images
Figure CN121831902A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of seismic event detection, in particular to a seismic event detection method based on multi-scale transformation of a seabed application scenario. BACKGROUND
[0002] Seismic event detection on the seabed is a key link for tsunami early warning, plate activity monitoring and deep-sea geophysical research, and its detection performance directly determines the response time of marine disaster emergency response and the depth of understanding of deep-sea geological activity. Compared with land seismic station network, the special extreme environment of the seabed puts forward more stringent technical requirements for the seismic monitoring system, and the core difficulties are concentrated in three constraints: the extreme communication constraints of extremely low water channel bandwidth, high delay and high energy consumption; the strong time-varying complex noise interference formed by the interweaving of ocean current disturbance, biological noise and ship sound source; and the real-time rigid demand of minutes or even seconds for tsunami early warning.
[0003] The current mainstream method still mainly follows the detection idea of land station, such as the energy ratio threshold method based on STA / LTA or simple multi-channel consistency discrimination. These methods face serious limitations in the seabed scenario, such as the fact that modern seabed seismographs are generally equipped with three-axis velocity meters, three-axis accelerometers and hydrophones, but existing systems often only use part of the channels, failing to take advantage of the complementary advantages of pressure waves and solid waves in propagation speed and frequency band characteristics, missing the opportunity to identify early P waves.
[0004] Among them, the limitations of the prior art at least include the following problems, the prior art is difficult to balance real-time and detection accuracy, and the multi-source sensor information is not fully utilized, the complementary advantages of three-axis velocity, three-axis acceleration and hydrophone signals are not fused, the opportunity to identify early P waves is missed, in addition, there is a lack of a collaborative architecture that adapts to the constraints of underwater acoustic communication, the prior art is prone to cause a large amount of invalid data to occupy the low-bandwidth, high-delay underwater acoustic channel, and to increase the burden of the shore-based system; at the same time, it is difficult to realize a fast closed loop from detection to verification to confirmation, the cross-verification chain of multiple stations takes too long, and in addition, an efficient feature extraction and intelligent discrimination scheme is not designed for the seabed environment, the detection algorithm has poor adaptability, and it is difficult to cope with the problems of strong time-varying noise and sensor performance drift. SUMMARY
[0005] In view of the deficiencies of the prior art, the present application provides a seismic event detection method based on multi-scale transformation of a seabed application scenario, which solves the problems of the prior art that real-time accuracy is difficult to balance, multi-source information is not fused, communication adaptation is poor, and closed loop adaptation is insufficient.
[0006] To achieve the above object, the application is implemented by the following technical solutions: a seismic event detection method based on multi-scale transformation of seabed application scenarios, comprising the following steps: collecting a plurality of channel original signals based on a seabed seismograph and performing multi-scale time-frequency transformation processing to construct a fusion feature vector; performing local event preliminary screening based on the fusion feature vector to generate an event package and report to a buoy node; in the buoy node, receiving all event packages uploaded by seabed seismographs within its jurisdiction and performing joint verification processing to obtain a weighted comprehensive confidence of each event package, generate a swarm event report, and upload to a shore-based processing center; in the shore-based processing center, receiving all swarm event reports uploaded by buoy nodes within its range and performing clustering processing to obtain a plurality of same events and analyze the global existence probability of the same events; determining whether the global existence probability of each same event is higher than a preset global existence probability threshold, and determining the same event higher than the preset global existence probability threshold as a real seismic event.
[0007] Further, the original signal includes: three-axis velocity signal, three-axis acceleration signal, hydrophone signal.
[0008] Further, the specific steps of constructing the fusion feature vector are as follows: preprocessing each channel original signal; performing S transformation on the preprocessed original signal of each channel; performing feature extraction processing on the original signal of each channel after S transformation to obtain an evaluation feature set of each channel original signal, including time-frequency energy, first arrival time difference, and average coherence coefficient; and constructing a fusion feature vector based on the evaluation feature set of each channel.
[0009] Further, the specific steps of generating an event package are as follows: inputting the fusion feature vector into a lightweight 1D-CNN model to output a seismic probability; and determining whether the seismic probability is higher than a preset seismic probability threshold; if the seismic probability is higher than the preset seismic probability threshold, determining that it is an effective trigger and generating an event package.
[0010] Further, the specific steps of obtaining the weighted comprehensive confidence of each event package are as follows: performing spatio-temporal consistency screening processing on each event package to obtain the theoretical P-wave travel time of each event package and determine whether it meets a preset first screening condition; performing multi-scale feature similarity clustering on each event package that meets the preset first screening condition to obtain the cosine similarity of any two event packages and determine whether it meets a preset second screening condition; performing cross-validation processing on each event package that meets the preset third screening condition; and generating the weighted comprehensive confidence of each event package based on each event package after cross-validation processing.
[0011] Furthermore, the specific steps to obtain several identical events are as follows: perform cross-cluster event association processing on the cluster event reports uploaded by each buoy node, and extract the time interval and spatial distance of the cluster event reports uploaded by each group of buoy nodes; if the time interval and spatial distance of the cluster event reports uploaded by each group of buoy nodes meet the preset clustering conditions, they are marked as identical events.
[0012] Furthermore, the specific steps for analyzing the global existence probability of each identical event are as follows: read the weighted comprehensive confidence scores corresponding to several independent reports from multiple clusters in each identical event; based on Bayesian global confidence fusion, perform statistical processing on the weighted comprehensive confidence scores corresponding to each independent report from each cluster in each identical event to obtain the global existence probability of each identical event.
[0013] The present invention has the following beneficial effects:
[0014] (1) This seismic event detection method based on multi-scale integrated transformation of seabed application scenarios achieves efficient unity of rapid response and accurate identification in seismic event detection through multi-level collaborative design and deep integration of multi-source information throughout the entire process. It collects raw signals of three-axis velocity, three-axis acceleration and hydrophone channel, and after preprocessing and S-transformation, constructs fused feature vectors to allow multi-dimensional perception data to fully play complementary roles. Then, it outputs the earthquake probability through a lightweight 1D-CNN model to complete the local initial screening and ensure the accuracy of trigger judgment. The spatiotemporal consistency screening, feature similarity clustering and cross-validation at the buoy node level are used to screen effective information and generate weighted comprehensive confidence. The shore-based processing center accurately determines real earthquake events through cross-group event association clustering and Bayesian global confidence fusion, which not only ensures detection speed but also greatly improves the recognition accuracy. At the same time, it adapts to the transmission requirements of underwater acoustic communication scenarios, making information transmission more efficient.
[0015] (2) The seismic event detection method based on multi-scale integrated transformation of seabed application scenarios has greatly improved the reliability of seismic event detection through refined layered processing. In the data processing stage, multi-scale time-frequency transformation and multi-dimensional feature extraction are combined to fully capture the key features of seismic signals and avoid judgment bias caused by single-dimensional analysis. The probability threshold judgment in the local initial screening stage, the multi-round screening verification of buoy nodes, and the global existence probability assessment of the shore-based center form a multi-level verification mechanism. At the same time, the simplified design of event packets and cluster event reports reduces the amount of data transmission while ensuring the integrity of key information, and adapts to the needs of different communication scenarios.
[0016] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description
[0017] Figure 1 This is a flowchart of the seismic event detection method based on multi-scale transformation in the seabed application scenario of the present invention.
[0018] Figure 2 This is a flowchart illustrating the specific steps involved in obtaining the weighted comprehensive confidence level of each event packet in the multi-scale transformation seismic event detection method based on seabed application scenarios of this invention. Detailed Implementation
[0019] Please see Figure 1 This invention provides a technical solution: a seismic event detection method based on multi-scale transformation of seabed application scenarios, comprising the following steps: acquiring raw signals from several channels (e.g., 7 channels) using a seabed seismograph, performing multi-scale time-frequency transformation processing, and constructing a fused feature vector; performing initial screening of local events based on the fused feature vector, generating event packets, and reporting them to buoy nodes; in the buoy nodes, receiving event packets uploaded by all seabed seismographs within their jurisdiction, performing joint verification processing, obtaining the weighted comprehensive confidence level of each event packet, generating a cluster event report, and uploading it to the shore-based processing center;
[0020] In the shore-based processing center, cluster event reports uploaded by all buoy nodes within its range are received and clustered to obtain several identical events. The global probability of these identical events is analyzed. It is then determined whether the global probability of each identical event is higher than a preset global probability threshold. Identified identical events with a global probability threshold as real earthquake events are then located using nonlinear methods (such as grid search or particle filtering). The final epicenter, time of occurrence, and magnitude are output, and an early warning or archive is triggered.
[0021] The raw signals include: triaxial velocity signals Triaxial acceleration signal Hydrophone signal .
[0022] Specifically, the steps for constructing the fused feature vector are as follows: Preprocess the original signal for each channel, i.e., to unify the analysis basis, integrate the acceleration signal to obtain the velocity estimate: And an exponentially weighted fusion method is used to suppress integral drift: The preprocessed raw signal of each channel is subjected to a (modified) S-transform (Generalized S-Transform, GST), i.e.: , where the window width function ,Pick This enables high time resolution in the low-frequency band and excellent frequency resolution in the high-frequency band, adapting to the characteristics of seismic P-wave.
[0023] Feature extraction is performed on the original signal of each channel after S-transformation to obtain the evaluation feature set of the original signal of each channel, including time-frequency energy, first-arrival time difference, and average coherence coefficient, i.e., in the P-wave sensitive frequency band. Internally, calculate the time-frequency energy of each channel: Simultaneously calculate: the P-wave first arrival time difference between the hydrophone and the vertical velocity channel (located by cross-correlation peaks); the average coherence coefficient among the six solid-state channels: ,in Based on the evaluation feature set of each channel, a fused feature vector is constructed, that is, a 13-dimensional fused feature vector is constructed: .
[0024] The specific steps for generating the event package are as follows: Input the fused feature vector into a lightweight 1D-CNN model, and output the earthquake probability, i.e. It then determines whether the earthquake probability is higher than a preset earthquake probability threshold; if the earthquake probability is higher than the preset earthquake probability threshold, i.e. If the default value is 0.85, it is considered a valid trigger, and an event packet is generated, including the device ID, trigger time, etc. Confidence level .
[0025] like Figure 2 As shown, the specific steps to obtain the weighted composite confidence level of each event packet are as follows: Perform spatiotemporal consistency screening on each event packet to obtain the theoretical P-wave travel time of each event packet, and determine whether it meets the preset first screening condition, i.e.: Let the first event packet be... The OBS location corresponding to each event packet is: Trigger time is Assuming the initial epicenter is located at the geometric center of the aircraft group Calculate the theoretical P-wave travel time: The first screening criterion is If the first preset screening condition is not met, it will be removed;
[0026] For each event packet that meets the preset first screening condition, perform multi-scale feature similarity clustering to obtain the cosine similarity between any two event packets, and determine whether it meets the preset second screening condition, that is: the event packet that meets the preset first screening condition corresponds to the first... Feature vectors of OBS Calculate the pairwise cosine similarity: And calculate the average similarity, the second screening criterion is and ;
[0027] For each event packet that meets the preset third screening criteria, a (hydrophone-solid wave) cross-validation process is performed, that is, each node is verified. Does it conform to the wavefront propagation direction (i.e., the closer to the epicenter, the better)? The smaller the value, the less likely it is to be a mismatch; those that do not match are removed. Based on each event packet after cross-validation, a weighted comprehensive confidence score is generated for each event packet, i.e.: .
[0028] This implementation scheme employs a meticulously designed, progressively layered approach to ensure robust support for each stage of seismic event detection. This guarantees both the comprehensiveness of information and the reliability of judgments. During preprocessing, acceleration integrals are converted to velocity and drift is suppressed, providing a unified analytical foundation for signals from different channels. The improved S-transform, with its specific window width setting, perfectly matches the characteristics of seismic P-waves, accurately capturing time variations in the low-frequency band and clearly distinguishing frequency differences in the high-frequency band, laying a high-quality foundation for subsequent feature extraction. The 13-dimensional fused feature vector integrates multiple key information types, including time-frequency energy, first-arrival time difference, and average coherence coefficient. It avoids single-dimensional judgments, comprehensively capturing the core features of seismic signals, thus providing a solid basis for subsequent analysis. Based on this, a lightweight 1D-CNN model outputs earthquake probabilities, combined with a default threshold trigger of 0.85. This ensures both the efficiency of the initial screening and the preliminary filtering of invalid signals. The generated event packets contain only key information such as device ID and trigger time, avoiding redundant data that consumes resources. The spatiotemporal consistency screening of buoy nodes removes abnormal event packets based on P-wave travel time. Feature similarity clustering ensures the consistency of features of valid events through cosine similarity. Hydrophone-solid wave cross-validation further verifies the rationality of the signals. The weighted comprehensive confidence score generated after multiple rounds of screening provides a clear quantitative standard for the credibility of each event packet, ensuring that no valid signals are missed and no interference information is misled, thus significantly improving the reliability of the detection results.
[0029] Specifically, the steps to obtain several identical events are as follows: Perform cross-cluster event correlation processing on the cluster event reports uploaded by each buoy node, and extract the time interval and spatial distance of the cluster event reports uploaded by each group of buoy nodes (a group consists of any two buoy nodes); if the time interval and spatial distance of the cluster event reports uploaded by each group of buoy nodes meet the preset clustering conditions, i.e., the time interval... Spatial distance If they are the same event, then they are marked as the same event.
[0030] The specific steps for analyzing the global existence probability of each identical event are as follows: First, read the weighted comprehensive confidence scores corresponding to several independent reports from multiple clusters for each identical event; second, based on Bayesian global confidence fusion, statistically process the weighted comprehensive confidence scores corresponding to each independent report from each cluster for each identical event to obtain the global existence probability of each identical event. Let a certain event be... Each cluster reports independently, with a confidence level of [value missing]. The global probability is: .
[0031] In this implementation plan, the accuracy of event determination is significantly improved by using association rules that fit the actual scenario and a scientific confidence fusion method. When associating cross-fleet events, the clustering conditions of time interval ≤30s and spatial distance ≤100km are used to accurately classify different fleet reports of the same earthquake event into one category, avoiding omissions or misjudgments. Subsequently, based on Bayesian global confidence fusion, the independent weighted comprehensive confidence of multiple fleets under the same event is integrated. It does not rely solely on the result of a single fleet, but performs statistical calculations based on the confidence of multiple sources, making the global probability of existence more convincing. The whole process not only ensures the accuracy of classifying the same event, but also provides a solid basis for the final earthquake event determination through multi-dimensional confidence integration, effectively avoiding the deviation that may occur in the judgment of a single node or fleet.
[0032] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.
[0033] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.
Claims
1. A seismic event detection method based on multi-scale transformation in seabed application scenarios, characterized in that, Includes the following steps: Based on the acquisition of raw signals from several channels by the seabed seismometer, and the multi-scale time-frequency transformation processing, a fused feature vector is constructed. Local events are initially screened based on fused feature vectors, event packets are generated, and reported to the buoy node; In the buoy node, event packets uploaded by all seabed seismometers within its jurisdiction are received, and joint verification processing is performed to obtain the weighted comprehensive confidence of each event packet, generate a cluster event report, and upload it to the shore-based processing center. In the shore-based processing center, cluster event reports uploaded by all buoy nodes within its range are received, clustered, and several identical events are obtained. The global probability of the existence of these identical events is then analyzed. Determine whether the global probability of each identical event is higher than a preset global probability threshold, and identify identical events that are higher than the preset global probability threshold as real earthquake events.
2. The seismic event detection method based on multi-scale integrated transformation of seabed application scenarios according to claim 1, characterized in that, The raw signals include: triaxial velocity signals, triaxial acceleration signals, and hydrophone signals.
3. The seismic event detection method based on multi-scale integrated transformation of seabed application scenarios according to claim 2, characterized in that, The specific steps for constructing the fused feature vector are as follows: Preprocess the raw signal for each channel; Perform an S-transform on the original signal of each preprocessed channel; Feature extraction processing is performed on the original signal of each channel after S-transformation to obtain the evaluation feature set of the original signal of each channel, including time-frequency energy, first arrival time difference, and average coherence coefficient; Based on the evaluation feature set of each channel, a fused feature vector is constructed.
4. The seismic event detection method based on multi-scale integrated transformation of seabed application scenarios according to claim 1, characterized in that, The specific steps for generating the event package are as follows: The fused feature vectors are input into a lightweight 1D-CNN model, which outputs the earthquake probability. And determine whether the earthquake probability is higher than the preset earthquake probability threshold; If the earthquake probability is higher than the preset earthquake probability threshold, it is determined to be a valid trigger and an event packet is generated.
5. The seismic event detection method based on multi-scale integrated transformation of seabed application scenarios according to claim 1, characterized in that, The specific steps to obtain the weighted overall confidence score for each event packet are as follows: For each event packet, a spatiotemporal consistency screening process is performed to obtain the theoretical P-wave travel time of each event packet, and it is determined whether the preset first screening condition is met. For each event packet that meets the preset first screening condition, perform multi-scale feature similarity clustering to obtain the cosine similarity between any two event packets, and determine whether the preset second screening condition is met. Cross-validation is performed on each event packet that meets the preset third screening criteria; Based on each event packet after cross-validation, a weighted overall confidence score is generated for each event packet.
6. The seismic event detection method based on multi-scale integrated transformation of seabed application scenarios according to claim 1, characterized in that, The specific steps to obtain several identical events are as follows: Perform cross-cluster event correlation processing on the cluster event reports uploaded by each buoy node, and extract the time interval and spatial distance of the cluster event reports uploaded by each group of buoy nodes; If the time interval and spatial distance of the cluster event reports uploaded by each group of buoy nodes meet the preset clustering conditions, they are marked as the same event.
7. The seismic event detection method based on multi-scale integrated transformation of seabed application scenarios according to claim 6, characterized in that, The specific steps for analyzing the global probability of each identical event are as follows: Read the weighted composite confidence level corresponding to several independent reports from multiple clusters for each identical event; Based on Bayesian global confidence fusion, the weighted comprehensive confidence scores corresponding to the independent reports of each cluster in each identical event are statistically processed to obtain the global existence probability of each identical event.