Illegal fishing early warning method and system based on underwater sound environment monitoring

Abstract

The invention discloses an illegal fishing early warning method and system based on underwater sound environment monitoring, and the method comprises the steps: determining an underwater acoustic equipment layout scheme according to the area of a to-be-monitored water area and the monitoring range of underwater acoustic equipment; therefore, underwater acoustic data of the to-be-monitored water area is obtained; sequentially performing sampling rate regularization processing, length cutting processing and digital filtering processing on the obtained underwater acoustic data of the water area to be monitored to obtain a signal to be identified; inputting the to-be-recognized signal into the sound environment recognition model to obtain a sound environment monitoring recognition result; performing fishing early warning according to the sound environment monitoring identification result; the sound environment recognition model comprises a feature extractor and a neural network classifier; the feature extractor is a depth auto-encoder composed of a plurality of convolution blocks and is used for extracting sound environment audio features from the to-be-identified signal; the sound environment monitoring identification result comprises the following conditions: no ship exists, normal sailing is carried out if a ship exists, and fishing activity is carried out if a ship exists.

Description

Technical Field

[0001] The present invention belongs to the technical fields of artificial intelligence and natural resource management fishery law enforcement, and particularly relates to an illegal fishing warning method and system based on underwater acoustic environment monitoring. Background Art

[0002] During the fishing moratorium, illegal fishing activities still occur in some river sections and lakes, and the concealment of illegal fishing activities is getting higher and higher. Especially at night without light conditions, the water surface visual range is extremely limited, and it is difficult to quickly and accurately detect illegal fishing activities through monitoring means such as optics and vision. The water area supervision and law enforcement is difficult, which brings great challenges to front-line law enforcement officers.

[0003] To further consolidate the achievements of the fishing moratorium and implement the fishing moratorium plan, it is necessary to further improve the monitoring ability of illegal fishing activities in key water areas during key time periods, timely dissuade the discovered illegal fishing behaviors, and protect the water area ecology and fishery resource safety. Summary of the Invention

[0004] Object of the Invention: To solve the technical problem that illegal fishing activities cannot be monitored relying on optical images at night and under poor light conditions, the present invention proposes an illegal fishing warning method and system based on underwater acoustic environment monitoring, and through underwater acoustic technical means, acoustic environment monitoring is carried out on key law enforcement water areas during important time periods.

[0005] Technical Solution: An illegal fishing warning method based on underwater acoustic environment monitoring includes the following steps:

[0006] Step 1: Determine the layout plan of underwater acoustic devices according to the area of the water area to be monitored and the monitoring range of underwater acoustic devices; according to the layout plan of underwater acoustic devices, obtain the underwater acoustic data of the water area to be monitored;

[0007] Step 2: Perform sampling rate regularization processing, length cutting processing, and digital filtering processing on the obtained underwater acoustic data of the water area to be monitored in sequence to obtain a signal to be recognized;

[0008] Step 3: Input the signal to be recognized into an acoustic environment recognition model to obtain an acoustic environment monitoring recognition result;

[0009] Step 4: Carry out fishing warning according to the acoustic environment monitoring recognition result;

[0010] Among them, the sound environment recognition model includes: a feature extractor and a neural network classifier; the feature extractor is a deep autoencoder composed of multiple convolutional blocks, which is used to extract sound environment audio features from the signal to be recognized, and the neural network classifier is used to classify the sound environment according to the sound environment audio features to obtain the sound environment monitoring and recognition result; the sound environment monitoring and recognition result includes: no vessels, vessels sailing normally, and vessels engaged in fishing activities.

[0011] Further, the following steps are also included between step 1 and step 2:

[0012] Perform gain control, filtering operation, and amplification processing on the obtained underwater acoustic data of the water area to be monitored in sequence to obtain optimized underwater acoustic data;

[0013] Step 2 is replaced by the following steps:

[0014] Perform sampling rate regularization processing, length cutting processing, and digital filtering processing on the optimized underwater acoustic data in sequence to obtain the signal to be recognized.

[0015] Further, the sound environment recognition model is trained according to the following steps:

[0016] Construct a complete encoder, where the complete encoder includes an encoding module and a decoding module; combine a vector quantization codebook and train the complete encoder through unsupervised training to minimize the loss function.

[0017] Take out the trained encoding module as the feature extractor in the sound environment recognition model, combine it with a neural network classifier for supervised training, and obtain the trained sound environment recognition model by minimizing the error between the predicted value and the label value.

[0018] The present invention discloses an illegal fishing warning system based on underwater sound environment monitoring, including:

[0019] An underwater acoustic data acquisition module, which is used to determine the layout scheme of underwater acoustic devices according to the area of the water area to be monitored and the monitoring range of underwater acoustic devices; and obtain the underwater acoustic data of the water area to be monitored according to the layout scheme of underwater acoustic devices.

[0020] A data processing module, which is used to perform sampling rate regularization processing, length cutting processing, and digital filtering processing on the obtained underwater acoustic data of the water area to be monitored in sequence to obtain the signal to be recognized;

[0021] An identification module, which is used to input the signal to be recognized into the sound environment recognition model to obtain the sound environment monitoring and recognition result;

[0022] A warning module, which is used to perform fishing warning according to the sound environment monitoring and recognition result;

[0023] Among them, the sound environment recognition model includes: a feature extractor and a neural network classifier; the feature extractor is a deep autoencoder composed of multiple convolutional blocks, which is used to extract sound environment audio features from the signal to be recognized, and the neural network classifier is used to classify the sound environment according to the sound environment audio features to obtain the sound environment monitoring and recognition result; the sound environment monitoring and recognition result includes: no vessels, vessels sailing normally, and vessels engaged in fishing activities.

[0024] Furthermore, a data optimization module is further included between the underwater acoustic data acquisition module and the data processing module;

[0025] The data optimization module is used to perform gain control, filtering operation and amplification processing on the underwater acoustic data of the water area to be monitored obtained in sequence to obtain the optimized underwater acoustic data;

[0026] Among them, the data processing module is used to perform sampling rate regularization processing, length cutting processing and digital filtering processing on the optimized underwater acoustic data in sequence to obtain the signal to be recognized.

[0027] Furthermore, the sound environment recognition model is trained according to the following steps:

[0028] Construct a complete encoder, the complete encoder includes an encoding module and a decoding module; combined with a vector quantization codebook, the complete encoder is trained by minimizing the loss function through unsupervised training;

[0029] Take out the trained encoding module as the feature extractor in the sound environment recognition model, and perform supervised training in combination with the neural network classifier. By minimizing the error between the predicted value and the label value, the trained sound environment recognition model is obtained.

[0030] Beneficial effects: Compared with the prior art, the present invention has the following advantages

[0031] (1) The propagation of sound waves underwater is not restricted by optical conditions, so it can be used for the monitoring of illegal fishing activities at night and under poor optical conditions;

[0032] (2) Compared with optical image means, the hardware system used in the acoustic-based monitoring method is simpler, so it is more convenient to deploy, has low device power consumption and lower system cost;

[0033] (3) The intelligent underwater sound environment anomaly monitoring combined with deep learning can automatically and intelligently identify illegal fishing activities in the water area and give early warnings, greatly reducing the work burden of front-line law enforcement officers. Description of the Drawings

[0034] Figure 1 It is a schematic diagram of the composition of an illegal fishing early warning system based on underwater sound environment monitoring proposed by the present invention;

[0035] Figure 2 It is a flowchart of the specific implementation method of an illegal fishing warning system based on underwater acoustic environment monitoring;

[0036] Figure 3 They are the key component modules of the system;

[0037] Figure 4 It is a schematic diagram of a layout method with strong engineering practicability;

[0038] Figure 5 It is a typical state classification directory block diagram for monitoring the acoustic environment in key waters during important time periods;

[0039] Figure 6 It is a workflow diagram for acoustic environment classification modeling and training. Specific implementation manners

[0040] The technical solution of the present invention will be further elaborated below in conjunction with the accompanying drawings.

[0041] Example 1:

[0042] Sound waves are the only signal form currently found that can propagate long distances underwater. During the navigation and operation of fishing boats, the tail propeller needs to rotate continuously underwater to generate propulsion force. During the propulsion process, the propeller interacts with the water body to generate noise, and the generated noise carries obvious ship physical characteristic information (number of blades, rotational speed, etc.). Therefore, by monitoring the changes in the underwater acoustic noise environment (referred to as the acoustic environment) in key waters during important time periods, illegal fishing activities can be discovered in a timely and accurate manner.

[0043] Therefore, this example proposes an illegal fishing warning method based on underwater acoustic environment monitoring, which uses the noise characteristics generated during the navigation and fishing operations of fishing boats, combined with artificial intelligence deep learning algorithms, to achieve effective warning of illegal fishing activities. As Figure 2 shown, it mainly includes the following steps:

[0044] Step 1: According to the area of the key monitoring waters and the parameters of the underwater acoustic equipment, determine the layout plan to ensure that the underwater acoustic data of the entire key monitoring waters can be collected; assume that the area of the key waters is D, and the shape is approximately circular. Since the hydrophone is isotropic, let the effective action distance of the hydrophone be R, then the action range is a circle with the buoy platform as the center point, and it can be arranged according to Figure 4 to carry out the layout. Use the underwater acoustic equipment to collect the underwater acoustic data during important time periods, and convert the acoustic data into electrical signals;

[0045] Step 2: Perform processing such as gain control, filtering, and amplification on the signals output by the acoustic data acquisition module to optimize the signal quality and reliability;

[0046] Step 3: Perform data conversion and length cutting on the optimized data, and perform preprocessing such as digital filtering on the cut data to obtain the signal to be recognized. This step can optimize the signal quality and improve the reliability. Among them, data conversion is to regularize the sampling rate of the optimized data to maintain a consistent sampling rate; length cutting is to cut each signal into an appropriate length for subsequent processing and storage, etc.; digital filtering is to filter out noise outside the target frequency band and other existing interferences, etc., to improve the signal-to-noise ratio of the sound environment monitoring signal.

[0047] Step 4: Input the signal to be recognized into the sound environment recognition model, complete the extraction of acoustic signal features, feature classification, and sound environment state recognition, obtain the sound environment monitoring recognition result, and judge whether there is illegal fishing operation in this monitoring water area according to the output sound environment monitoring recognition result. If so, feedback this sound environment monitoring recognition result to the monitoring command center.

[0048] As Figure 6 shown, the sound environment recognition model involved in this embodiment is trained according to the following steps, including:

[0049] Construct an end-to-end sound environment recognition model. Aiming at the actual needs of underwater sound environment modeling and classification for the early warning task of illegal fishing behavior in important event segments of key waters and the characteristics of difficult processing of underwater sound signals, the constructed sound environment recognition model includes: a feature extractor and a feature classifier; the feature extractor is the encoding module of a deep autoencoder, which consists of multiple convolutional blocks and is used to extract the audio features of the sound environment monitor. The feature classifier is used to implement the specific task of sound environment classification and recognition.

[0050] When training the encoder for the feature extractor, a complete encoder needs to be constructed to achieve this. The complete encoder includes an encoding module and a decoding module. Combining with a vector quantization codebook, the loss function is minimized through unsupervised training to achieve audio compression and noise reduction. Take out the obtained encoding module after training as the feature extractor of the sound environment data, and perform supervised training in combination with a neural network classifier to minimize the error between the predicted value and the label value.

[0051] Construct a training data set; according to different sound environment types, perform data cleaning and annotation on the processed information to form data sets of different sound environment types for classification modeling and model training of sound environment monitoring. As Figure 5 shown, the typical state classification catalog of the monitored sound environment during important time periods in specific key waters includes: no vessel state, vessel with normal navigation state, and vessel with fishing activity state.

[0052] Example 2:

[0053] As Figure 3As shown in the figure, this embodiment proposes an illegal fishing warning system based on underwater acoustic environment monitoring, which mainly includes two parts: a wet-end module and a dry-end module. The wet-end module refers to the acoustic sensors deployed underwater. The acoustic sensors are hydrophones mounted on a buoy platform. By deploying the hydrophones in the key monitoring target waters, they are used to collect underwater acoustic data during important time periods. The hydrophones can convert underwater acoustic wave signals into electrical signals. The dry-end module includes a signal conditioning module, a data processing module, an acoustic environment recognition module, a communication module, and a control module.

[0054] The signal conditioning module is connected to the backend of the hydrophone and is used to perform processing such as gain control, filtering, and amplification on the signals output by the acoustic data acquisition module to optimize the signal quality and reliability.

[0055] The data processing module is used to perform functions such as data conversion, segmentation cutting, digital filtering, and storage on the signals output by the signal conditioning module to obtain the warning signals to be recognized. Among them, data conversion is to perform sampling rate regularization processing on the optimized data to maintain a consistent sampling rate; length cutting is to cut each signal into an appropriate length for subsequent processing and storage, etc.; digital filtering is to filter out the noise outside the target frequency band and other existing interferences to improve the signal-to-noise ratio of the acoustic environment monitoring signals.

[0056] The acoustic environment recognition module is used to input the warning signals to be recognized into the acoustic environment recognition model to obtain the warning results and realize the real-time monitoring of the acoustic environment in key waters.

[0057] In this embodiment, a communication module and a control module are also included. The communication module realizes the transmission of collected data and control instructions through wireless communication. The control module is used to control the overall operation of the system, including data acquisition, data transmission, data processing parameter configuration, acoustic environment monitoring and recognition, decision-making, etc. The control system turns on the acoustic sensors during important monitoring time periods to continuously monitor and collect the acoustic signal data in key waters.

[0058] The present invention relies on underwater acoustic technology means to realize the acoustic environment monitoring of key law enforcement waters during important time periods, and can be used for the monitoring of illegal fishing activities at night and when the optical conditions are poor. The hardware system is simple and can automatically and intelligently identify illegal fishing activities in waters and give warnings.

Claims

1. An illegal fishing warning method based on underwater soundscape monitoring, characterized in that: Including the following steps: Step 1: Determine the underwater acoustic device layout plan according to the area of the water area to be monitored and the monitoring range of the underwater acoustic device; Obtain the underwater acoustic data of the water area to be monitored according to the underwater acoustic device layout plan; Step 2: Sequentially perform sampling rate regularization processing, length cutting processing, and digital filtering processing on the obtained underwater acoustic data of the water area to be monitored to obtain the signal to be recognized; Step 3: Input the signal to be recognized into the sound environment recognition model to obtain the sound environment monitoring recognition result; Step 4: Perform fishing warning according to the sound environment monitoring recognition result; Wherein, the sound environment recognition model includes: a feature extractor and a neural network classifier; the feature extractor is a deep autoencoder composed of multiple convolutional blocks, which is used to extract sound environment audio features from the signal to be recognized, and the neural network classifier is used to perform sound environment classification according to the sound environment audio features to obtain the sound environment monitoring recognition result; the sound environment monitoring recognition result includes: no vessel, vessel sailing normally, and vessel engaged in fishing activities.

2. The illegal fishing warning method based on underwater acoustic environment monitoring according to claim 1, wherein: Between Step 1 and Step 2, the following steps are also included: Sequentially perform gain control, filtering operation, and amplification processing on the obtained underwater acoustic data of the water area to be monitored to obtain optimized underwater acoustic data; Step 2 is replaced with the following steps: Sequentially perform sampling rate regularization processing, length cutting processing, and digital filtering processing on the optimized underwater acoustic data to obtain the signal to be recognized.

3. The illegal fishing warning method based on underwater acoustic environment monitoring according to claim 1, characterized in that: The sound environment recognition model is trained according to the following steps: Construct a complete encoder, the complete encoder includes an encoding module and a decoding module; combined with a vector quantization codebook, the complete encoder is trained by minimizing the loss function through unsupervised training; Take out the trained encoding module as the feature extractor in the sound environment recognition model, and perform supervised training in combination with the neural network classifier. By minimizing the error between the predicted value and the label value, the trained sound environment recognition model is obtained.

4. An illegal fishing warning system based on underwater acoustic environment monitoring, characterized in that: Including: An underwater acoustic data acquisition module, which is used to determine the underwater acoustic device layout plan according to the area of the water area to be monitored and the monitoring range of the underwater acoustic device; Obtain the underwater acoustic data of the water area to be monitored according to the underwater acoustic device layout plan; A data processing module, which is used to sequentially perform sampling rate regularization processing, length cutting processing, and digital filtering processing on the obtained underwater acoustic data of the water area to be monitored to obtain the signal to be recognized; An identification module, which is used to input the signal to be recognized into the sound environment recognition model to obtain the sound environment monitoring recognition result; A warning module, which is used to perform fishing warning according to the sound environment monitoring recognition result; Wherein, the sound environment recognition model includes: a feature extractor and a neural network classifier; the feature extractor is a deep autoencoder composed of multiple convolutional blocks, which is used to extract sound environment audio features from the signal to be recognized, and the neural network classifier is used to perform sound environment classification according to the sound environment audio features to obtain the sound environment monitoring recognition result; the sound environment monitoring recognition result includes: no vessel, vessel sailing normally, and vessel engaged in fishing activities.

5. The illegal fishing warning system based on underwater acoustic environment monitoring according to claim 4, characterized in that: A data optimization module is also included between the underwater acoustic data acquisition module and the data processing module; The data optimization module is used to perform gain control, filtering operation, and amplification processing on the obtained underwater acoustic data of the water area to be monitored in sequence, so as to obtain optimized underwater acoustic data; Among them, the data processing module is used to perform sampling rate regularization processing, length cutting processing, and digital filtering processing on the optimized underwater acoustic data in sequence to obtain the signal to be recognized.

6. The illegal fishing warning system based on underwater acoustic environment monitoring according to claim 4, characterized in that: The acoustic environment recognition model is trained according to the following steps: Construct a complete encoder, where the complete encoder includes an encoding module and a decoding module; combine a vector quantization codebook and train the complete encoder by minimizing the loss function through unsupervised training; Take out the trained encoding module as the feature extractor in the acoustic environment recognition model, combine it with a neural network classifier for supervised training, and obtain the trained acoustic environment recognition model by minimizing the error between the predicted value and the label value.

Images