A method and device for identifying fish sound signals based on single-channel hydrophone

Abstract

The invention discloses a method and device for identifying fish sound signals based on a single-channel hydrophone, belonging to the field of water sound signal and pattern recognition. A double-threshold activity segment detection algorithm is designed to extract the sound fish signals of the water sound activity segment, and the water sound activity segment signal is decomposed into three-scale wavelet packets to extract the energy characteristics and singular value characteristics of 8 wavelet packet coefficients. Then, a separability judgment method of cluster separation degree is used to find feature quantities with large separability to form a feature vector. Finally, a machine learning method based on a BP neural network is designed to learn sample data, and a category judgment is made using the principle of minimum root mean square error of the output vector, thereby realizing the identification of the sound fish and obtaining a higher recognition rate.

Description

Technical Field

[0001] The invention relates to the field of combining water acoustic signals with pattern recognition algorithms, and in particular to a method and device for recognizing fish acoustic signals based on a single-channel hydrophone. Background Art

[0002] There are complex underwater acoustic signals in the marine environment, which are a mixture of information such as seabed geological activities, marine biological activities, and hydrological activities. Shallow-water dolphins, humpback whales, totoaba, shrimps and many other marine organisms emit low-frequency acoustic signals during their life activities. The vocalization frequencies of Atlantic cod for feeding and reproduction are mainly concentrated at 80~500Hz, the lowest frequency of whale songs can be below 10Hz, humpback whales can emit acoustic signals of about 30Hz when hitting the water, and blue whales in the South Indian Ocean emit acoustic signals with energy concentrated at 20Hz~40Hz. The acoustic signals emitted by these biological activities can help marine biologists explore the biological secrets of the habitats, distribution, and living habits of marine organisms such as fish and marine mammals.

[0003] Hydrophones are widely used sensors in ocean acoustic detection. They have high sensitivity and can convert sound wave signals in the water into electrical signals. Hydroacoustic detection equipment using hydrophones is carried on marine observation platforms such as buoys and buoys. They can observe the radiation noise of underwater targets, seabed geological activities, cetacean activities, etc., and record hydroacoustic signals.

[0004] The identification of fish based on multi-channel hydrophone signals involves large amounts of data processing and low utilization of effective signal information. In addition, there are many hydrophones, which are inconvenient to arrange and cumbersome to operate. In addition, single-domain features contain one-sided information. Time domain features lack frequency domain information, and frequency domain features cannot characterize the timing of actions. Therefore, the present invention integrates time-frequency processing, wavelet analysis, and other methods, and effectively realizes the recognition of the sound signals of various fish through an activity segment detection method, a time-frequency domain feature extraction method, and a machine learning method based on a neural network. Summary of the invention

[0005] The purpose of the present invention is to address the deficiencies of the prior art and propose a method and device for identifying fish acoustic signals based on a single-channel hydrophone. The double-threshold activity segment detection algorithm is designed to extract the fish acoustic signals of the water acoustic activity segment, and the water acoustic activity segment signal is decomposed into three-scale wavelet packets to extract the energy characteristics and singular value characteristics of 8 wavelet packet coefficients. Then, the separability judgment method of cluster separation degree is used to find the feature vector composed of feature quantities with large separability. Finally, a machine learning method based on BP neural network is designed to learn the sample data, and the category judgment is made using the principle of minimum root mean square error of the output vector, so as to realize the identification of the sound-making fish and obtain a higher recognition rate.

[0006] The object of the present invention is achieved through the following technical solutions: In the first aspect, the present invention provides a method for identifying fish sound signals based on a single-channel hydrophone, the method comprising the following steps:

[0007] (1) Obtaining the activity water sound signals of fish in the area through hydrophones;

[0008] (2) The activity underwater acoustic signal is divided into several segments through the first sliding window, and the features are extracted and compared with the first threshold to obtain a single threshold activity segment detection signal. If it is determined to be active, 1 is output, and if it is determined to be inactive, 0 is output;

[0009] (3) taking the single-threshold activity segment detection signal through the second sliding window frame, performing cumulative calculation on the activity segment output, and comparing the calculation result with the second threshold to obtain a dual-threshold activity segment detection signal;

[0010] (4) Select the water acoustic activity segment data according to the time sequence of the activity segment data points of the double-threshold activity segment detection signal, perform wavelet transform, and then use the cluster separation method to calculate the class distance of the three-scale wavelet packet decomposition coefficient characteristics of the water acoustic activity segment signal of each sound-making fish, as well as the class distance of the same feature quantity of the water acoustic activity segment signals of different sound-making fish; select the feature vector according to the principle of maximum separability and use principal component analysis to perform dimensionality reduction processing to obtain the feature vectors of the water acoustic signals of different sound-making fish, and construct training samples;

[0011] (5) Train the BP neural network based on the training samples, and use the trained network to determine the action category based on the sound signals of the vocalizing fish.

[0012] Furthermore, a single-channel hydrophone was arranged in the experimental anechoic pool to collect the current basic environmental noise. Then different species of vocal fish were put into the anechoic pool in turn, and the fish vocal signals were observed and recorded for the same length of time. The mean of the time corresponding to the fish vocal signal and the basic environmental noise at each sampling data point was calculated, and the active water acoustic signals of multiple groups of fish greater than the signal-to-noise ratio threshold were obtained to construct a data set. The sliding window method was used for segmentation, and the signal data within the sliding window length was extracted to obtain the characteristic quantity characterizing the intensity of the water acoustic signal.

[0013] Furthermore, the first threshold value is specifically set as follows: calculating the similarity between each data point in the window and the fish occurrence signal in the data set, extracting data points whose similarity is greater than the similarity threshold, and calculating the mean of these data points as the first threshold value.

[0014] Furthermore, the selection of the second sliding window length is related to the duration of fish actions. Based on the occurrence signals of fish actions in the data set, the mean duration of fish activities in the data set is calculated and set as the sliding window length, and the second threshold is set to 0.8 times the length of the first threshold sliding window.

[0015] Furthermore, the absolute average AMA of the sound signal of the sounding fish in the sliding window in the time domain is extracted as the threshold feature quantity characterizing the signal activity intensity. If the accumulated value of multiple consecutive active segment outputs is greater than the second threshold, the signal segment is considered to be an active segment signal.

[0016] Furthermore, in step (4), the water acoustic activity segment data is subjected to a three-scale wavelet packet transform to obtain eight wavelet packet decomposition coefficients, and a sliding window method is used to extract features from the fixed-length data in the sliding window to extract energy features and singular value features.

[0017] Furthermore, the input matrix of the training samples consists of the feature vectors of each action, each row is a sample, and the expected output matrix is ​​the action label, each row is a label vector, and different labels correspond to different actions.

[0018] In a second aspect, the present invention also provides a device for identifying fish sound signals based on a single-channel hydrophone, comprising a memory and one or more processors, wherein the memory stores executable code, and when the processor executes the executable code, the method for identifying fish sound signals based on a single-channel hydrophone is implemented.

[0019] In a third aspect, the present invention further provides a computer-readable storage medium having a program stored thereon, and when the program is executed by a processor, the method for identifying fish sound signals based on a single-channel hydrophone is implemented.

[0020] In a fourth aspect, a computer program product includes a computer program, and when the computer program is executed by a processor, the method for identifying fish sound signals based on a single-channel hydrophone is implemented.

[0021] The advantages of the present invention are:

[0022] (1) Wavelet packet decomposition is used to process and extract features of the underwater acoustic signals of vocal fish, splitting the single-channel signal into different frequency bands, and then extracting the signal features of different frequency bands, thereby splitting the one-dimensional single-channel signal into multi-dimensional signals, and obtaining a feature space containing time domain information and frequency domain information;

[0023] (2) Compared with the multi-channel fish acoustic signal time domain feature method, the method has a higher information utilization rate and a smaller number of channels; it only needs to repeatedly obtain the sound signals of fish in the test anechoic pool through fewer sensors, and the method and model set by the present invention can realize the recognition of fish signals based on a smaller amount of data, and the dual threshold method can realize a deeper level of feature judgment, which can effectively reduce the interference of background noise and improve the recognition accuracy.

[0024] (3) The dual threshold method activity segment detection link is added to the signal feature extraction to improve the extraction efficiency of the action activity segment feature and avoid the invalid operation of the inactive segment signal feature extraction. Compared with the single threshold method activity segment detection, the misjudgment phenomenon of the action activity segment detection signal is reduced.

[0025] (4) The cluster separation analysis method is used to select appropriate feature vectors. Compared with the non-screening method, this improves the recognition accuracy of the classifier and reduces the number of training iterations of neural network machine learning. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0027] Figure 1 This is a flow chart of the recognition of the underwater acoustic signal of vocalizing fish according to the present invention.

[0028] Figure 2 This is a flow chart of the detection of active segment signals of fish underwater acoustic signals using the dual threshold method of the present invention.

[0029] Figure 3 This is a training flow chart of the BP neural network matrix used in the present invention.

[0030] Figure 4 The present invention provides a structural diagram of a device for identifying fish sound signals based on a single-channel hydrophone. DETAILED DESCRIPTION

[0031] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with the accompanying drawings and implementation examples. It should be understood that the specific implementation examples described herein are only used to explain the present invention and are not used to limit the present invention.

[0032] like Figure 1 As shown, the present invention provides a method for identifying underwater acoustic signals of sound-emitting fish, and its specific implementation steps are as follows:

[0033] Step (1) Collecting water sound signals of different types of fish:

[0034] A single-channel hydrophone is arranged in the test anechoic pool to collect the current basic environmental noise. Then different types of vocal fish are placed in the anechoic pool in turn. The single-channel hydrophone is used to observe and record the fish vocal signals for the same length of time. The signal-to-noise ratio of the fish vocal signal and the basic environmental noise at each sampling data point is calculated. The active hydroacoustic signals of multiple groups of fish greater than the signal-to-noise ratio threshold are obtained to construct a data set. In the experiment, the vocal signals of a fish are recorded repeatedly, and multiple groups of hydroacoustic signals are collected to form sufficient training samples. In the hydroacoustic signal conditioning hardware circuit in the single-channel hydrophone, an instrument amplifier is used to perform a first-stage amplification processing on the differential input hydrophone, and a second-order voltage-controlled voltage source type low-pass filter is used to attenuate signals below 20KHz. In order to avoid the influence of the mains on the collected signal, a double-T double-follower notch filter is used to notch the 50Hz power frequency noise, and an integrated operational amplifier is used to perform a second-stage amplification processing on the filtered signal. And the sound signal intensity of different vocal fish is normalized.

[0035] Step (2) uses a double threshold method to detect the activity segment of the sound signal of the sound-emitting fish, such as Figure 2 As shown:

[0036] The sliding window method is adopted with a sampling frequency of 1000Hz. Taking into account the fluctuation degree of the characteristic quantity, the fixed data point length is set to 16. At this time, the sliding window time length is 16ms, and the collected fish acoustic signals are divided into several segments. The absolute average AMA of the sounding fish acoustic signals in the sliding window is extracted as a measure of the intensity of the acoustic signal characterizing the signal activity.

[0037]

[0038] Where N represents the number of sliding window data points, is the hydroacoustic signal intensity of data point i;

[0039] The value AMA is compared with the first threshold value set according to the signal strength to obtain a single threshold activity segment detection signal. If it is greater than or equal to the first threshold value, it is judged as 1. If it is less than the first threshold value, it is judged as 0. The sliding window method is used for the obtained single threshold activity segment detection signal, and the single threshold activity segment detection signal within the window length is accumulated and calculated, and the calculation result is compared with the second threshold value. If it is greater than or equal to the second threshold value, it is judged as 1, which is regarded as an action activity. If it is less than the second threshold value, it is judged as 0, which is regarded as an action inactivity.

[0040] The first threshold is set specifically as follows: based on the training sample obtained in step (1), the similarity between each data point in the window and the training sample is calculated, data points whose similarity is greater than the similarity threshold are extracted, and the mean of these data points is calculated as the first threshold; the setting of the first threshold can avoid the interference of underwater acoustic background noise.

[0041] The second threshold is set specifically as follows: the selection of the sliding window length is related to the duration of the fish action. Based on the training samples obtained in step (1), the mean duration of the fish activity in the training samples is calculated and set as the sliding window length. The second threshold is set to 0.8 times the length of the first threshold sliding window.

[0042] Step (3) obtaining the activity segment signal of the sound-emitting fish water sound signal:

[0043] The activity segment signal of the sounding fish water acoustic signal is extracted from the data sequence of the start and end of the action of the sounding fish water acoustic signal detection signal obtained in step (2). The Daubechies wavelet system db5 wavelet basis is used to perform 3-scale wavelet packet decomposition on the activity segment signal. According to the principle of maximum separability, 32 eigenvalues ​​with strong separability are selected to form a eigenvector with 31 rows and 1 column. A sliding window with 32 data points is used to extract the variance feature within the window. ;

[0044]

[0045] As the energy characteristic, the energy intensity is calculated as follows :

[0046]

[0047] In addition, the singular value features of the vectors in the sliding window are extracted;

[0048] Step (4) Selection of feature vector:

[0049] In the above steps, 8 feature quantities were obtained for each action. The cluster separation method was used to calculate the class distance of the three-scale wavelet packet decomposition coefficient characteristics of each sounding fish water acoustic activity segment signal and the class distance of the same feature quantity of the water acoustic activity segment signal of different sounding fish. According to the principle of maximum separability, 32 eigenvalues ​​with strong separability were selected to form a feature vector with 31 rows and 1 column. The principal component analysis method was used to reduce the dimension of the 32-dimensional feature vector, and the first 4 principal components were selected as the feature vector. The energy coefficients and singular value coefficients of the wavelet packet decomposition tree nodes (3, 1) and (3, 2) were selected to form a feature vector with one row and four columns. The feature vector was divided into a training vector and a test vector.

[0050] Step (5) Identification of fish water sound signals based on BP neural network:

[0051] like Figure 3As shown, according to the number of actions to be classified, the sample data for neural network training is constructed, including the input matrix and the expected output matrix, specifically: the feature vector of each action forms a neural network training input matrix with multiple rows and four columns. And set labels for different actions, that is, the expected output vector, and the expected output matrix is ​​a matrix with multiple rows and three columns composed of these labels. The training sample of the neural network is composed of the training input matrix and the expected output matrix. According to the dimension of the constructed feature vector and the dimension of the expected output vector, the BP neural network structure is designed as a three-layer network structure: including an input layer with 4 nodes, a hidden layer with 5 nodes and an output layer with 3 nodes. The segmented learning method is adopted to set the allowed global error and the maximum number of allowed iterations to train the sample data. After the training is completed, two weight matrices and two threshold vectors of the neural network are obtained, which can be represented as the trained neural network. The test vector is sequentially input into the neural network trained in step (5) to obtain the neural network calculation result, and the root mean square error between the expected output vectors of all action labels is calculated, and the action judgment is made according to the minimum root mean square error. The recognition accuracy is used to measure the recognition results of the BP neural network classifier. The recognition accuracy is the number of accurately recognized samples of the same label divided by the number of test samples of the same label.

[0052] Corresponding to the aforementioned embodiment of a method for identifying fish sound signals based on a single-channel hydrophone, the present invention also provides an embodiment of a device for identifying fish sound signals based on a single-channel hydrophone.

[0053] See also Figure 4 An embodiment of the present invention provides a device for identifying fish sound signals based on a single-channel hydrophone, comprising a memory and one or more processors, wherein the memory stores executable code, and when the processor executes the executable code, it is used to implement a method for identifying fish sound signals based on a single-channel hydrophone in the above embodiment.

[0054] An embodiment of a device for identifying fish sound signals based on a single-channel hydrophone provided by the present invention can be applied to any device with data processing capabilities, and the device with data processing capabilities can be a device or apparatus such as a computer. The device embodiment can be implemented through software, or through hardware or a combination of software and hardware. Taking software implementation as an example, as a device in a logical sense, it is formed by the processor of any device with data processing capabilities in which it is located reading the corresponding computer program instructions in the non-volatile memory into the internal memory for execution. From a hardware perspective, if Figure 4 As shown, it is a hardware structure diagram of a device having data processing capability for identifying fish sound signals based on a single-channel hydrophone provided by the present invention, except Figure 4In addition to the processor, memory, network interface, and non-volatile memory shown, any device with data processing capabilities in which the apparatus in the embodiments is located may also include other hardware, generally based on the actual functions of the device with data processing capabilities, which will not be described in detail.

[0055] The implementation process of the functions and effects of each unit in the above-mentioned device is specifically described in the implementation process of the corresponding steps in the above-mentioned method, and will not be repeated here.

[0056] For the device embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The device embodiment described above is only schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of the present invention. Ordinary technicians in this field can understand and implement it without paying creative work.

[0057] An embodiment of the present invention further provides a computer-readable storage medium having a program stored thereon. When the program is executed by a processor, a method for identifying fish sound signals based on a single-channel hydrophone in the above embodiment is implemented.

[0058] The computer-readable storage medium may be an internal storage unit of any device with data processing capability described in any of the aforementioned embodiments, such as a hard disk or a memory. The computer-readable storage medium may also be an external storage device of any device with data processing capability, such as a plug-in hard disk, a smart media card (SMC), an SD card, a flash card, etc. equipped on the device. Furthermore, the computer-readable storage medium may also include both an internal storage unit and an external storage device of any device with data processing capability. The computer-readable storage medium is used to store the computer program and other programs and data required by any device with data processing capability, and may also be used to temporarily store data that has been output or is to be output.

[0059] The present invention also provides a computer program product, including a computer program. When the computer program is executed by a processor, the method for identifying fish sound signals based on a single-channel hydrophone is implemented.

[0060] The above embodiments are used to illustrate the present invention rather than to limit the present invention. Any modification and change made to the present invention within the spirit of the present invention and the protection scope of the claims shall fall within the protection scope of the present invention.

Claims

1. A method for identifying fish sound signals based on a single-channel hydrophone, characterized in that: The method comprises the following steps: (1) Obtaining the activity water sound signals of fish in the area through hydrophones; (2) The activity underwater acoustic signal is divided into several segments through the first sliding window, and the features are extracted and compared with the first threshold to obtain a single threshold activity segment detection signal. If it is determined to be active, 1 is output, and if it is determined to be inactive, 0 is output; (3) taking the single-threshold activity segment detection signal through the second sliding window frame, performing cumulative calculation on the activity segment output, and comparing the calculation result with the second threshold to obtain a dual-threshold activity segment detection signal; (4) Select the water acoustic activity segment data according to the time sequence of the activity segment data points of the double-threshold activity segment detection signal, perform wavelet transform, and then use the cluster separation method to calculate the class median distance of the three-scale wavelet packet decomposition coefficient characteristics of the water acoustic activity segment signal of each sound-making fish, as well as the class distance of the same feature quantity of the water acoustic activity segment signals of different sound-making fish; select the feature vector according to the principle of maximum separability and use principal component analysis to perform dimensionality reduction processing to obtain the feature vectors of the water acoustic signals of different sound-making fish, and construct training samples; (5) Train the BP neural network based on the training samples, and use the trained network to determine the action category based on the sound signals of the vocalizing fish.

2. The method for identifying fish sound signals based on a single-channel hydrophone according to claim 1, characterized in that: A single-channel hydrophone was arranged in the experimental anechoic pool to collect the current basic environmental noise. Then different kinds of vocal fish were put into the anechoic pool in turn, and the fish vocal signals were observed and recorded for the same length of time. The mean of the time corresponding to the fish vocal signal and the basic environmental noise at each sampling data point was calculated, and the active water acoustic signals of multiple groups of fish greater than the signal-to-noise ratio threshold were obtained to construct a data set. The sliding window method was used to segment the data, and the signal data within the sliding window length was extracted to obtain the characteristic quantity characterizing the intensity of the water acoustic signal.

3. The method for identifying fish sound signals based on a single-channel hydrophone according to claim 2, characterized in that: The first threshold is set specifically as follows: calculate the similarity between each data point in the window and the fish signal in the data set, extract the data points whose similarity is greater than the similarity threshold, and calculate the mean of these data points as the first threshold.

4. The method for identifying fish sound signals based on a single-channel hydrophone according to claim 2, characterized in that: The selection of the second sliding window length is related to the duration of fish actions. Based on the occurrence signals of fish actions in the data set, the mean duration of fish activity in the data set is calculated and set as the sliding window length, and the second threshold is set to 0.8 times the length of the first threshold sliding window.

5. The method for identifying fish sound signals based on a single-channel hydrophone according to claim 1, characterized in that: The absolute average AMA of the sound signal of the sounding fish in the sliding window in the time domain is extracted and compared with the first threshold to obtain a single-threshold active segment detection signal. If the accumulated value of multiple consecutive active segment outputs is greater than the second threshold, the signal is considered to be an active segment signal.

6. The method for identifying fish sound signals based on a single-channel hydrophone according to claim 1, characterized in that: In step (4), the water acoustic activity segment data is subjected to a three-scale wavelet packet transform to obtain eight wavelet packet decomposition coefficients. The sliding window method is used to extract features from the fixed-length data in the sliding window to extract energy features and singular value features.

7. The method for identifying fish sound signals based on a single-channel hydrophone according to claim 1, characterized in that: The input matrix of the training sample consists of the feature vectors of each action, each row is a sample, and the expected output matrix is ​​the action label, each row is a label vector, and different labels correspond to different actions.

8. A device for identifying fish sound signals based on a single-channel hydrophone, comprising a memory and one or more processors, wherein the memory stores executable code, characterized in that: When the processor executes the executable code, a method for identifying fish sound signals based on a single-channel hydrophone as described in any one of claims 1 to 7 is implemented.

9. A computer-readable storage medium having a program stored thereon, characterized in that: When the program is executed by a processor, a method for identifying fish sound signals based on a single-channel hydrophone as described in any one of claims 1 to 7 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for identifying fish sound signals based on a single-channel hydrophone as described in any one of claims 1 to 7 is implemented.

Images