Fish feeding behavior identification method, device and equipment based on acoustic signal multi-feature representation, medium and product

Abstract

The invention discloses a fish feeding behavior recognition method, device and equipment based on acoustic signal multi-feature representation, a medium and a product, and relates to the technical field of underwater acoustic signal processing and machine learning. The method comprises the following steps: firstly, acquiring a fish ingestion sound sample signal, and dividing according to a preset time window to obtain sound sample signals of a plurality of time windows; respectively extracting statistical features and image representation features of the sound sample signals of each time window, and constructing a first training sample set and a second training sample set; then constructing a multi-feature fusion model; and training the multi-feature fusion model by using the first training sample set and the second training sample set to obtain a trained multi-feature fusion model, and performing fish feeding behavior identification. According to the method, the statistical features and the image representation features of the fish feeding sound signals are extracted to construct the multi-dimensional acoustic feature vectors, and the multi-feature fusion model is adopted to realize accurate identification of the fish feeding state in the culture pond.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of underwater acoustic signal processing and machine learning, in particular to a fish feeding behavior recognition method and device based on multi-feature representation of acoustic signals, equipment, medium and product. BACKGROUND

[0002] In industrial aquaculture, accurately identifying the feeding behavior of fish is of great significance for optimizing feeding strategies, improving feed utilization, and reducing breeding costs. Traditional feeding behavior recognition methods mainly rely on manual observation or simple acoustic monitoring. Existing acoustic monitoring-based methods mainly use a single type of acoustic feature, which cannot fully exploit the rich information in acoustic signals, resulting in low recognition accuracy. SUMMARY

[0003] The purpose of the present application is to provide a fish feeding behavior recognition method, device, equipment, medium and product based on multi-feature representation of acoustic signals, to improve the accuracy of fish feeding behavior recognition.

[0004] To achieve the above purpose, the present application provides the following solutions.

[0005] In a first aspect, the present application provides a fish feeding behavior recognition method based on multi-feature representation of acoustic signals, comprising: obtaining fish feeding sound sample signals; dividing the fish feeding sound sample signals according to a preset time window to obtain a plurality of time window sound sample signals; extracting statistical features of each time window sound sample signal to construct a first training sample set; extracting image representation features of each time window sound sample signal to construct a second training sample set; constructing a multi-feature fusion model; the multi-feature fusion model includes a PCA model and a frequency spectrum graph model; respectively inputting the statistical features into the PCA model and the image representation features into the frequency spectrum graph model, and outputting the fusion feeding behavior classification results of the PCA model and the frequency spectrum graph model, training the multi-feature fusion model using the first training sample set and the second training sample set to obtain a trained multi-feature fusion model; using the trained multi-feature fusion model to recognize fish feeding behavior.

[0006] Optionally, after dividing the fish feeding sound sample signals according to a preset time window to obtain a plurality of time window sound sample signals, the method further comprises: bandpass filtering each time window sound sample signal to obtain a bandpass filtered sound sample signal of each time window. subtracting the ambient noise baseline from the filtered sound sample signal of each time window to obtain an ambient noise subtracted sound sample signal of each time window; performing adaptive filtering on the ambient noise subtracted sound sample signal of each time window to obtain an adaptive filtered sound sample signal of each time window; performing wavelet denoising on the adaptive filtered sound sample signal of each time window to obtain a wavelet denoised sound sample signal of each time window; performing amplitude normalization on the wavelet denoised sound sample signal of each time window.

[0007] Optionally, the statistical features of the sound sample signal of each time window are extracted, specifically including: extracting time domain statistical features of the sound sample signal of each time window; extracting frequency domain statistical features of the sound sample signal of each time window; performing dimensionality reduction on the time domain statistical features and the frequency domain statistical features of the sound sample signal of each time window by using a principal component analysis method to obtain statistical features of the sound sample signal of each time window.

[0008] Optionally, image representation features of the sound sample signal of each time window are extracted, specifically including: respectively converting the sound sample signal of each time window into a two-dimensional spectrogram representation; respectively performing size standardization and flattening on the two-dimensional spectrogram of each time window to obtain an image representation feature of each time window.

[0009] Optionally, the sound sample signal of each time window is converted into a two-dimensional spectrogram representation, specifically including: performing frame processing on the sound sample signal of each time window; performing windowing processing and FFT transformation on each frame of sound sample signal of each time window to obtain an amplitude spectrum of each frame of sound sample signal of each time window; calculating a log power spectrum of each frame of sound sample signal of each time window according to the amplitude spectrum of each frame of sound sample signal of each time window; constructing a two-dimensional spectrogram representation of each time window based on the log power spectrum of each frame of sound sample signal of each time window.

[0010] Optionally, the PCA model includes a random forest model, a gradient boosting tree model, a support vector machine model, and a multilayer perception model; The feeding behavior classification result of the PCA model is obtained by fusing the feeding behavior classification results of the random forest model, the gradient boosting tree model, the support vector machine model, and the multilayer perception model by using a software voting mechanism; The spectrogram model is a CNN neural network model.

[0011] In a second aspect, the present application provides a fish feeding behavior recognition device based on acoustic signal multi-feature representation, characterized in that the fish feeding behavior recognition device based on acoustic signal multi-feature representation is applied to the fish feeding behavior recognition method based on acoustic signal multi-feature representation, and the fish feeding behavior recognition device based on acoustic signal multi-feature representation comprises a sound collection module, a signal processing module and a control module. The sound collection module is configured to acquire a fish feeding sound signal; the fish feeding sound signal is a fish feeding sound sample signal or a fish feeding sound to-be-tested signal. The signal processing module is configured to divide the fish feeding sound sample signal according to a preset time window to obtain a plurality of time window sound sample signals. The control module is configured to extract statistical features of the sound sample signals of each time window to construct a first training sample set; extract image representation features of the sound sample signals of each time window to construct a second training sample set; construct a multi-feature fusion model; the multi-feature fusion model comprises a PCA model and a spectrogram model; the statistical features are taken as the input of the PCA model, the image representation features are taken as the input of the spectrogram model, and the fusion feeding behavior classification results of the PCA model and the spectrogram model are taken as the output; the first training sample set and the second training sample set are used to train the multi-feature fusion model to obtain a trained multi-feature fusion model; and the trained multi-feature fusion model is used to recognize the fish feeding behavior.

[0012] In a third aspect, the present application provides a computer device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, and the processor executes the computer program to implement the fish feeding behavior recognition method based on acoustic signal multi-feature representation.

[0013] In a fourth aspect, the present application provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the fish feeding behavior recognition method based on acoustic signal multi-feature representation.

[0014] In a fifth aspect, the present application provides a computer program product, comprising a computer program, and the computer program is executed by a processor to implement the fish feeding behavior recognition method based on acoustic signal multi-feature representation.

[0015] According to the specific embodiments provided by the present application, the present application has the following technical effects.

[0016] This application provides a method, apparatus, device, medium, and product for recognizing fish feeding behavior based on multi-feature representation of acoustic signals. The application first acquires fish feeding sound sample signals; divides the fish feeding sound sample signals into multiple time windows according to a preset time window; then extracts statistical features of the sound sample signals in each time window to construct a first training sample set; extracts image representation features of the sound sample signals in each time window to construct a second training sample set; then constructs a multi-feature fusion model, which includes a PCA model and a spectrogram model; and trains the multi-feature fusion model using the first and second training sample sets to obtain a trained multi-feature fusion model; finally, it uses the trained multi-feature fusion model to recognize fish feeding behavior. This application extracts statistical and image representation features of fish feeding sound signals to construct a multi-dimensional acoustic feature vector, and then, based on the characteristics of noise changes in factory aquaculture environments, uses a multi-feature fusion model to accurately identify the feeding state of fish in aquaculture ponds. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating a method for recognizing fish feeding behavior based on multi-feature representation of acoustic signals, provided as an embodiment of this application.

[0019] Figure 2 This is a schematic diagram of a method for recognizing fish feeding behavior based on multi-feature representation of acoustic signals, provided in an embodiment of this application.

[0020] Figure 3 This is a structural diagram of a fish feeding behavior recognition device based on multi-feature representation of acoustic signals, provided in an embodiment of this application.

[0021] Figure 4 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0022] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments of the present application. Based on the embodiments of the present application, all the other embodiments obtained by a person of ordinary skill in the art without creative work fall within the scope of the present application.

[0023] The above-mentioned purposes, features and advantages of the present application can be more apparent and easy to understand. The present application will be described in further detail below with reference to the drawings and specific embodiments.

[0024] The present application aims to provide a fish feeding behavior recognition method, device, equipment, medium and product based on acoustic signal multi-feature representation, so as to improve the fish feeding behavior recognition accuracy.

[0025] The main reasons for the low recognition accuracy of traditional acoustic monitoring technology are as follows: 1. Single feature limitation: The existing technology mainly uses a single type of acoustic feature (such as time domain or frequency domain feature), which cannot fully exploit the rich information in the acoustic signal, resulting in low recognition accuracy.

[0026] 2. Incomplete feature representation: The traditional method fails to effectively combine the statistical characteristics and image representation characteristics of the acoustic signal, and the feature expression ability is limited.

[0027] 3. Poor model generalization ability: A single model is prone to overfitting and has poor adaptability in different breeding environments.

[0028] 4. Lack of real-time performance: The existing system has a complex processing flow, which is difficult to meet the real-time monitoring needs of industrialized breeding.

[0029] 5. Lack of standardized process: The complete technical solution from sound collection to result output is missing, and the coordination efficiency between modules is low.

[0030] Therefore, there is an urgent need for a fish feeding behavior recognition scheme that can fully utilize the multi-dimensional features of acoustic signals, has high recognition accuracy and good generalization ability.

[0031] In one exemplary embodiment, a fish feeding behavior recognition method based on acoustic signal multi-feature representation is provided, as shown in Figure 1 and Figure 2 The method comprises the following steps 101-107.

[0032] Step 101: Acquire fish feeding sound sample signals.

[0033] Step 102: Divide the fish feeding sound sample signals according to a preset time window to obtain a plurality of time window sound sample signals.

[0034] Step 103, extracting the statistical features of the sound sample signal of each time window to construct a first training sample set.

[0035] Step 104, extracting the image representation features of the sound sample signal of each time window to construct a second training sample set.

[0036] Step 105, constructing a multi-feature fusion model; the multi-feature fusion model includes a PCA model and a spectrum graph model.

[0037] Step 106, respectively taking the statistical features as the input of the PCA model, taking the image representation features as the input of the spectrum graph model, taking the fusion feeding behavior classification results of the PCA model and the spectrum graph model as the output, training the multi-feature fusion model by using the first training sample set and the second training sample set to obtain a trained multi-feature fusion model.

[0038] Step 107, using the trained multi-feature fusion model to recognize the fish feeding behavior.

[0039] The above steps 101-107 are a fish feeding behavior recognition method based on a pre-trained model and multi-feature representation of acoustic signals. Through the integrated learning model pre-trained, the statistical features (PCA dimension reduction features) and the image representation features (spectrum graph features) of the acoustic signals are fused to realize the non-contact, high-precision real-time inference recognition of the fish feeding sound in the time window.

[0040] In another exemplary embodiment, the above step 102 can be implemented by the following process.

[0041] The sound segment is segmented by using the fixed time window technology, the window size is 5 seconds, and there is no overlap Each time window is independently preprocessed and inferred, and the algorithm output delay is 5 seconds.

[0042] In another exemplary embodiment, the above step 102 further includes: Step 201, performing band-pass filtering on the sound sample signal of each time window to obtain the band-pass filtered sound sample signal of each time window; Step 202, subtracting the environmental noise baseline from the filtered sound sample signal of each time window to obtain the environmental noise subtracted sound sample signal of each time window; Step 203, performing adaptive filtering on the environmental noise subtracted sound sample signal of each time window to obtain the adaptive filtered sound sample signal of each time window; Step 204, wavelet denoising is performed on the adaptive filtered sound sample signal of each time window to obtain a wavelet denoised sound sample signal of each time window; Step 205, amplitude normalization is performed on the wavelet denoised sound sample signal of each time window.

[0043] In another exemplary embodiment, the above steps 201-205 are performed on the sound data in each 5s time window, and the preprocessing includes: application of a band-pass filter (100Hz-8000Hz) to remove high and low frequency noise ambient noise reduction: subtracting the pre-established ambient noise baseline from the collected sound signal adaptive filtering: dynamically adjusting the filter parameters according to the ambient noise changes wavelet denoising: using wavelet transform to further remove residual noise normalization: amplitude normalization of the denoised signal to eliminate the influence of sensor sensitivity differences.

[0044] In another exemplary embodiment, high-sensitivity digital hydrophones are used to collect underwater sound data, with a sampling rate of 128kHz and a bit width of 16bit; Before formal collection, 1 minute of ambient noise is collected as a baseline; Record the spectral characteristics and statistical properties of the ambient noise; Establish an ambient noise template for noise reduction in subsequent signal processing.

[0045] In another exemplary embodiment, the sound sample signal of each time window is processed as follows: 1. Statistical feature extraction (PCA feature) A1, extract the time domain statistical features of the acoustic signal: mean, variance, skewness, kurtosis.

[0046] A2, extract the frequency domain statistical features: spectral centroid, spectral bandwidth, spectral roll-off point. The above two steps extract a total of 78-dimensional feature vectors.

[0047] A3, apply principal component analysis (PCA) for dimension reduction, retaining the first 11 principal component features. The measured results show that the contribution rate of the first 11 principal components is >95%.

[0048] 2. Image representation feature extraction (spectrogram feature) B1, convert the one-dimensional acoustic signal to a two-dimensional spectrogram representation.

[0049] B2, spectrogram parameters: frame length 1024 points, frame shift 512 points, FFT point number 2048.

[0050] B3, the spectrogram size is standardized to 64x64 pixels, and after flattening, a 4096-dimensional feature vector is obtained.

[0051] The image representation feature extraction process can be implemented using the following steps: C1, frame the one-dimensional acoustic signal: frame length 1024 samples, frame shift 512 samples (50% overlap). The total number of frames is calculated as follows: , total number of frames, signal length, frame length, frame shift.

[0052] C2, windowing: Hann window, formula: . Wherein, sample number, the amplitude value (or "window coefficient") of the window function at the nth sample point, total number of samples (window length) contained by the window function.

[0053] C3, FFT transform: 1025 frequency components (amplitude spectrum) are obtained for each frame.

[0054] C4, calculate the log power spectrum, formula: , wherein is a very small positive number to prevent negative infinity from occurring in the result.

[0055] C5, size normalization: 128k, 5s audio, corresponding frame number = 128000*5 / 512 = 1250, then the original spectrum size = 1250*1025; after bicubic interpolation resampling, the target size is 64*64 pixels.

[0056] C6, flatten the feature vector: flatten the 64*64 matrix in row priority order to a 4096-dimensional feature vector, and normalize it to the [-3, 20] dB range.

[0057] In another exemplary embodiment, the above-mentioned step 103 can be replaced by the following steps 301-303.

[0058] Step 301, extract the time-domain statistical features of the sound sample signal in each time window; Step 302, extract the frequency-domain statistical features of the sound sample signal in each time window; Step 303, use principal component analysis to reduce the dimensionality of the time-domain statistical features and frequency-domain statistical features of the sound sample signal in each time window, and obtain the statistical features of the sound sample signal in each time window.

[0059] In another exemplary embodiment, the above-mentioned step 104 can be replaced by the following steps 401 and 402.

[0060] Step 401, respectively, convert each time window sound sample signal into a two-dimensional spectrum map representation.

[0061] Step 402, respectively, size standardization and flattening is carried out on each time window two-dimensional spectrum map, and the image representation characteristics of each time window are obtained.

[0062] The specific implementation of step 401 is as follows: Frame processing is performed on the sound sample signal of each time window; Window processing and FFT transformation are performed on each frame sound sample signal of each time window, and the amplitude spectrum of each frame sound sample signal of each time window is obtained; According to the amplitude spectrum of each frame sound sample signal of each time window, the log power spectrum of each frame sound sample signal of each time window is calculated; Based on the log power spectrum of each frame sound sample signal of each time window, a two-dimensional spectrum map representation of each time window is constructed.

[0063] In another exemplary embodiment, the feature fusion obtained by the embodiment includes: The 11-dimensional PCA feature is spliced with the 4096-dimensional spectrum map feature to obtain a 4107-dimensional fusion feature vector, and the feature vector is standardized by using the Z-score standardization method.

[0064] In another exemplary embodiment, in steps 105 and 106, the pre-trained model is loaded and inferred first.

[0065] For the data collected in the 5s time window, after preprocessing, the multi-feature fusion model is input, and the model is trained and inferred, including: D1, PCA model pre-training stage: use a large amount of labeled fish feeding sound data to train the PCA model, and the PCA model includes four base learners: Random Forest model (Random Forest): 100 trees, maximum depth 10.

[0066] Gradient Boosting tree model (Gradient Boosting): learning rate 0.1, 100 iterations.

[0067] Support Vector Machine model (SVM): RBF kernel function, C=10, gamma=0.001.

[0068] Neural network (MLP): hidden layer (100, 50), activation function ReLU.

[0069] D2, reasoning stage: directly load pre-training model parameters, real-time inference on pre-processed audio data, and result fusion by soft voting mechanism.

[0070] D3, training stage.

[0071] Prepare data: statistical features of audio segments, image representation features, Train multi-feature fusion model, wherein 11 statistical features are used to train 4 classifiers (random forest, gradient boosting, SVM, neural network); image representation features are used to train 1 CNN neural network model.

[0072] The two models (PCA model and spectrum graph model) in the multi-feature fusion model respectively output the probabilities corresponding to the "strong, medium, weak, and no" feeding intensity.

[0073] Test different weight combinations to find the best weights for the PCA model and the spectrum graph model.

[0074] Calculate the fusion result: final probability = PCA weight x PCA probability + spectrum graph weight x spectrum graph probability.

[0075] In another exemplary embodiment, the above step 107 identifies the feeding intensity based on the trained multi-feature fusion model, and outputs the classification result: no feeding, weak feeding, medium feeding, and strong feeding. Based on the 5-second time window processing mechanism, the system response delay is stable at 5 seconds.

[0076] Based on the same inventive concept, the embodiments of the present application also provide an acoustic signal multi-feature representation based fish feeding behavior identification device for implementing the above-mentioned acoustic signal multi-feature representation based fish feeding behavior identification method. The problem-solving implementation scheme provided by the device is similar to the implementation scheme described in the above method, so the specific limitations in one or more acoustic signal multi-feature representation based fish feeding behavior identification device embodiments provided below can be referred to the limitations of the acoustic signal multi-feature representation based fish feeding behavior identification method in the above, which will not be repeated here.

[0077] In one exemplary embodiment, an acoustic signal multi-feature representation based fish feeding behavior identification device is provided, as shown in Figure 3 including: a sound collection module, a signal processing module and a control module; The sound collection module is used to acquire fish feeding sound signals; the fish feeding sound signals are fish feeding sound sample signals or fish feeding sound to-be-tested signals; The signal processing module is used to divide the fish feeding sound sample signals according to a preset time window to obtain a plurality of time window sound sample signals. The control module is configured to extract statistical features of the sound sample signal of each time window to construct a first training sample set, extract image representation features of the sound sample signal of each time window to construct a second training sample set, construct a multi-feature fusion model, the multi-feature fusion model comprising a PCA model and a spectrum graph model, input the statistical features into the PCA model, input the image representation features into the spectrum graph model, and input the fusion foraging behavior classification results of the PCA model and the spectrum graph model as output, train the multi-feature fusion model by using the first training sample set and the second training sample set to obtain a trained multi-feature fusion model, and use the trained multi-feature fusion model to identify the fish foraging behavior.

[0078] The device of the above embodiment is applied to a factory farming scene, and can realize quasi-real-time identification of fish foraging behavior, output results and save historical data.

[0079] In another exemplary embodiment, the above-mentioned sound collection module comprises: Hydrophone: high-sensitivity piezoelectric hydrophone, frequency response range 10Hz-20kHz, sensitivity -180dB re 1V / μPa, working depth 0-10 meters, protection level IP68.

[0080] Pre-amplifier: gain 40dB, input impedance 10MΩ, output impedance 100Ω, noise factor <3dB, supporting differential input to suppress common-mode interference.

[0081] Waterproof housing: 316L stainless steel material, pressure resistance 10bar, equipped with water-tight connector, supporting long-term deployment in seawater.

[0082] In another exemplary embodiment, the above-mentioned signal processing module comprises: Front-end analog filter: 6th order Butterworth band-pass filter, passband 100Hz-20kHz, passband ripple <0.5dB, stopband attenuation >40dB.

[0083] ADC: 16-bit high-speed analog-to-digital converter, sampling rate 128kHz, dynamic range 96dB, total harmonic distortion plus noise (THD+N) <-90dB.

[0084] Digital signal processor: 32-bit floating-point DSP, main frequency 400MHz, integrated hardware FFT accelerator, supporting real-time 512-point FFT operation.

[0085] De-noising algorithm: spectral subtraction plus wavelet threshold de-noising.

[0086] In another exemplary embodiment, the control module adopts a main chip of Ruiyi Micro RK3576, which meets the following indicators.

[0087] Main processor: ARM Cortex-A72 processor, frequency greater than 2GHz, supporting floating point acceleration.

[0088] AI inference unit: integrated NPU, computing power > 5TOPS, supporting INT8 / FP16 inference.

[0089] Memory: 4GB RAM + 32GB SD card or EMMC.

[0090] In another exemplary embodiment, the device further comprises a display module, which adopts a 10-inch industrial-grade TFT liquid crystal screen with a resolution of 1024x768 and is provided with a salt mist-proof shell.

[0091] In another exemplary embodiment, the device further comprises a communication module, which comprises: WiFi module: transmission rate > 100Mbps.

[0092] Ethernet: 100Mbps Ethernet, one RJ45 standard interface.

[0093] In another exemplary embodiment, the device further comprises a power management module, which comprises: a 12V / 20Ah lithium battery pack, a charging management submodule and a power consumption management submodule, wherein the cycle life of the 12V / 20Ah lithium battery pack is > 2000 times, the maximum charging current in the charging management submodule is 5A, the power consumption management submodule supports battery power supply and mains power supply, the working power consumption is < 15W, supports sleep mode, and the sleep power consumption is < 500mW.

[0094] In another exemplary embodiment, the working process of the device comprises an initialization process and a real-time identification process.

[0095] The initialization process comprises the following steps E1-E4.

[0096] E1, system startup: after the device is powered on, automatically perform hardware self-checking and software initialization, load preset identification parameters and threshold settings.

[0097] E2, pre-trained model loading: load the pre-trained integrated learning model parameters from the memory, including the weight parameters of random forest, gradient boosting tree, support vector machine and neural network.

[0098] E3, network connection: establish a communication connection with the server.

[0099] E4, calibration (optional): verify the accuracy of the system using standard test signals.

[0100] The real-time identification process includes the following steps F1-F5.

[0101] F1, sound collection: hydrophone continuously collects sound signals in the culture environment with a time window of 5 seconds F2, signal preprocessing: real-time filtering and denoising processing is performed on each 5-second window data F3, feature calculation: PCA features and spectrum features are calculated F4, pre-trained model inference: directly load pre-trained model parameters, input fusion features into pre-trained integrated model for real-time inference classification, and the final processing delay is 5 seconds.

[0102] F5, result output: Local display: display the current feeding intensity on the device screen (one result per window, i.e., update the fish feeding intensity every 5 seconds).

[0103] Data transmission: upload the results to the server through Ethernet / WiFi.

[0104] Control other equipment (optional): automatically adjust the feeding operation of the feeding equipment according to the feeding behavior recognition result.

[0105] In an exemplary embodiment, a computer device, which can be a server or a terminal, can have an internal structure as shown in Figure 4 The computer device includes a processor, a memory, an input / output interface (I / O), and a communication interface. The processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals through network connection. The computer program is executed by the processor to implement a fish feeding behavior recognition method based on multi-feature representation of acoustic signals.

[0106] Those skilled in the art can understand, Figure 4The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components. In an exemplary embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps in the above method embodiments.

[0107] In an exemplary embodiment, a computer readable storage medium is provided, storing a computer program, which is executed by a processor to implement the steps in the above method embodiments.

[0108] In an exemplary embodiment, a computer program product is provided, including a computer program, which is executed by a processor to implement the steps in the above method embodiments.

[0109] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties.

[0110] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when the computer program is executed, the processes of the above-mentioned embodiments of the methods can be included. Any reference to a memory, a database or other medium used in the embodiments provided in the present application can include at least one of a non-volatile and a volatile memory. The non-volatile memory can include a read-only memory (ROM), a magnetic tape, a floppy disk, a flash memory, an optical storage, a high-density embedded non-volatile memory, a resistive random access memory (ReRAM), a magnetoresistive random access memory (MRAM), a ferroelectric random access memory (FRAM), a phase change memory (PCM), a graphene memory, etc. The volatile memory can include a random access memory (RAM) or an external cache memory, etc. As an illustration but not limitation, the RAM can be in various forms, such as a static random access memory (SRAM) or a dynamic random access memory (DRAM), etc.

[0111] The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a blockchain, etc., without being limited thereto. The processor involved in the embodiments provided in the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., without being limited thereto.

[0112] The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of the technical features in the above embodiments are not described, however, as long as the combinations of the technical features do not exist contradictory, they should be considered as the scope of the present application.

[0113] The principles and implementation modes of the present application are described by using specific examples in the present application. The above embodiments are only used to help understand the method and its core idea of the present application; meanwhile, for those skilled in the art, according to the idea of the present application, the specific implementation mode and application range can be changed. In conclusion, the content of the present application should not be understood as a limitation.

Claims

1. A fish feeding behavior recognition method based on multi-feature representation of acoustic signals, characterized in that, The method comprises the following steps: acquiring fish feeding sound sample signals; dividing the fish feeding sound sample signals according to preset time windows to obtain multiple time window sound sample signals; extracting statistical features of the sound sample signals of each time window to construct a first training sample set; extracting image representation features of the sound sample signals of each time window to construct a second training sample set; constructing a multi-feature fusion model; the multi-feature fusion model comprises a PCA model and a frequency spectrum model; respectively taking the statistical features as the input of the PCA model, taking the image representation features as the input of the frequency spectrum model, taking the fusion feeding behavior classification results of the PCA model and the frequency spectrum model as the output, training the multi-feature fusion model by using the first training sample set and the second training sample set to obtain a trained multi-feature fusion model; using the trained multi-feature fusion model to identify the fish feeding behavior.

2. The fish feeding behavior recognition method based on multi-feature representation of acoustic signals according to claim 1, characterized in that, After the fish feeding sound sample signals are divided according to preset time windows to obtain multiple time window sound sample signals, the method further comprises the following steps: performing band-pass filtering on the sound sample signals of each time window to obtain the band-pass filtered sound sample signals of each time window; subtracting the environmental noise baseline from the filtered sound sample signals of each time window to obtain the environmental noise subtracted sound sample signals of each time window; performing adaptive filtering on the environmental noise subtracted sound sample signals of each time window to obtain the adaptive filtered sound sample signals of each time window; performing wavelet denoising on the adaptive filtered sound sample signals of each time window to obtain the wavelet denoised sound sample signals of each time window; performing amplitude normalization on the wavelet denoised sound sample signals of each time window. 3.The fish feeding behavior recognition method based on multi-feature representation of acoustic signals according to claim 1, characterized in that, The method for extracting the statistical features of the sound sample signals of each time window comprises the following steps: extracting time domain statistical features of the sound sample signals of each time window; extracting frequency domain statistical features of the sound sample signals of each time window; performing dimension reduction on the time domain statistical features and the frequency domain statistical features of the sound sample signals of each time window by using principal component analysis to obtain the statistical features of the sound sample signals of each time window.

4. The fish feeding behavior recognition method based on multi-feature representation of acoustic signals according to claim 1, characterized in that, The method for extracting the image representation features of the sound sample signals of each time window comprises the following steps: respectively converting the sound sample signals of each time window into two-dimensional frequency spectrum representations; respectively performing size standardization and flattening on the two-dimensional frequency spectrum of each time window to obtain the image representation features of each time window.

5. The fish feeding behavior recognition method based on multi-feature representation of acoustic signals according to claim 4, characterized in that, The method for converting the sound sample signals of each time window into two-dimensional frequency spectrum representations comprises the following steps: performing frame processing on the sound sample signals of each time window; performing windowing processing and FFT transformation on the sound sample signals of each frame of each time window to obtain the amplitude spectrum of the sound sample signals of each frame of each time window; calculating the log power spectrum of the sound sample signals of each frame of each time window according to the amplitude spectrum of the sound sample signals of each frame of each time window; constructing the two-dimensional frequency spectrum representation of each time window based on the log power spectrum of the sound sample signals of each frame of each time window.

6. The fish feeding behavior recognition method based on multi-feature representation of acoustic signals according to claim 1, characterized in that, The PCA model comprises a random forest model, a gradient boosting tree model, a support vector machine model and a multilayer perception model. The feeding behavior classification result of the PCA model is fused with the feeding behavior classification results of the random forest model, the gradient boosting tree model, the support vector machine model and the multilayer perception model by using a software voting mechanism. The spectrum graph model is a CNN neural network model.

7. An apparatus for fish feeding behavior recognition based on multi-feature representation of acoustic signals, characterized in that, The fish feeding behavior recognition device based on the multi-feature representation of acoustic signals is applied to the fish feeding behavior recognition method based on the multi-feature representation of acoustic signals according to any one of claims 1-6, and the fish feeding behavior recognition device based on the multi-feature representation of acoustic signals comprises a sound collection module, a signal processing module and a control module. The sound collection module is used to acquire fish feeding sound signals; the fish feeding sound signals are fish feeding sound sample signals or fish feeding sound signals to be tested. The signal processing module is used to divide the fish feeding sound sample signals according to a preset time window to obtain a plurality of time window sound sample signals. The control module is used to extract statistical features of the sound sample signals of each time window to construct a first training sample set; extract image representation features of the sound sample signals of each time window to construct a second training sample set; construct a multi-feature fusion model; the multi-feature fusion model comprises a PCA model and a spectrum graph model; the statistical features are taken as the input of the PCA model, the image representation features are taken as the input of the spectrum graph model, and the fusion feeding behavior classification results of the PCA model and the spectrum graph model are taken as the output; the first training sample set and the second training sample set are used to train the multi-feature fusion model to obtain a trained multi-feature fusion model; and the trained multi-feature fusion model is used to recognize fish feeding behaviors.

8. A computer device comprising: A memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the fish feeding behavior recognition method based on the multi-feature representation of acoustic signals according to any one of claims 1-6.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the fish feeding behavior recognition method based on the multi-feature representation of acoustic signals according to any one of claims 1-6.

10. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the fish feeding behavior recognition method based on the multi-feature representation of acoustic signals according to any one of claims 1-6.

Images