Fish Respiration Monitoring Methods and Systems

Abstract

This invention provides a method and system for monitoring fish respiration, belonging to the field of artificial intelligence technology. The method includes: acquiring fish video data; extracting images from the fish video data; using a hybrid cascaded network model to perform real-time inference on each frame of video to obtain motion bounding boxes for the movement of the fish mouth and gills in each frame; performing spatiotemporal feature extraction on the motion bounding boxes to obtain time-series motion signals of the fish mouth and gills; determining the fish's respiratory frequency value based on the time-series motion signals; and determining the fish's pressure level based on the fish's respiratory frequency value. Compared to existing technologies, the method and system provided by this invention, through the combination of a hybrid cascaded network model and time-series motion signals, achieve accurate monitoring of fish respiration under the influence of water ripples.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and more specifically to a method and system for monitoring fish respiration. Background Technology

[0002] In current aquaculture, the assessment of fish health typically relies on manual observation or traditional biosensors. These methods are often inefficient, lack accuracy, and are easily affected by environmental factors. With the development of computer vision technology, non-contact monitoring using video images has become possible. Optical flow, as an effective motion estimation method, has been used in various biological behavior analysis scenarios. However, traditional optical flow (such as Farneback) depends on image gradient changes, and is prone to losing motion information in low-texture areas or when fish are swimming rapidly. Summary of the Invention

[0003] The purpose of this invention is to provide a method and system for monitoring fish respiration, which improves the accuracy of monitoring fish respiration.

[0004] To achieve the above objectives, embodiments of the present invention provide a method for monitoring fish respiration, comprising:

[0005] Acquire fish video data;

[0006] Extract an image from the fish's video data;

[0007] A hybrid cascaded network model is used to perform real-time inference on each frame of video to obtain the motion bounding box of the fish mouth and gills in each frame.

[0008] Spatiotemporal feature extraction is performed on the motion bounding box to obtain the time-series motion signals of the fish mouth and gills;

[0009] The fish's respiratory rate value is determined based on the time-series motion signal;

[0010] The fish's body pressure level is determined based on the fish's respiratory rate.

[0011] Optionally, fish video data can be acquired, including:

[0012] Gaussian filtering is used to perform low-frequency filtering on the acquired raw video stream;

[0013] The median filtering method is used to process the original video stream after low-frequency filtering;

[0014] The original video stream after processing is adjusted using a multispectral channel equalization method;

[0015] The contrast-limited adaptive histogram equalization method is used to enhance the adjusted original video stream.

[0016] Optionally, it also includes:

[0017] The fish video is split frame by frame to obtain the fish image;

[0018] The head region of the fish in the fish image is labeled;

[0019] The labeled fish images are subjected to at least one of the following processing methods: random rotation, dynamic scaling, fogging simulation, and motion blur.

[0020] The processed fish images are combined into a training dataset.

[0021] The initial model was constructed using the ConvNeXt network structure as the backbone network.

[0022] Determine the loss function of the initial model according to formulas (1) to (3):

[0023] FL(p)=-α(1-p) γ log(p), (1)

[0024] BCE(p)=-log(p), (2)

[0025] Loss(p)=FL(p)+BCE(p), (3)

[0026] Where α is the balance factor, p is the prediction probability of the initial model, γ is the focus factor, and Loss(p) is the loss function;

[0027] The initial model is trained using the dataset to obtain the hybrid cascaded network model.

[0028] Optionally, spatiotemporal feature extraction is performed on the motion bounding box to obtain the time-series motion signals of the fish mouth and gills, including:

[0029] A multi-scale feature pyramid is used to extract the spatiotemporal motion features between motion bounding boxes of every two frames;

[0030] An iterative update mechanism is used to calculate the optical flow field components based on the aforementioned spatiotemporal motion characteristics;

[0031] The motion intensity characterization is calculated based on the optical flow field components.

[0032] A time-series signal is generated based on the motion intensity characterization at a preset sampling rate;

[0033] A time alignment mechanism is used to compensate for phase deviation inference of the time series signal by the model;

[0034] The time series signal is synchronized based on the phase deviation inference result to obtain the time series motion signal.

[0035] Optionally, an iterative update mechanism is used to calculate the optical flow field components based on the spatiotemporal motion characteristics, including:

[0036] A median filter is used to eliminate abnormal optical flow components in the optical flow field.

[0037] Optionally, the motion intensity characterization is calculated based on the optical flow field components, including:

[0038] The exercise intensity characterization is calculated using formula (4):

[0039]

[0040] in, For the motion intensity characterization of the i-th sampling point, u i v i denoted as different sub-components in the optical flow field component of the i-th sampling point, and λ is the statistical quantile.

[0041] Optionally, determining the fish's respiratory frequency value based on the time-series motion signal includes:

[0042] The time-series motion signal is preprocessed sequentially using a fourth-order Butterworth low-pass filter and a Symlet wavelet function.

[0043] A sliding window dynamic peak detection algorithm is used to determine respiratory cycle feature points based on the preprocessed time-series motion signal.

[0044] The weighted average of the intervals between adjacent peaks is determined based on the respiratory cycle characteristic points;

[0045] The fish's respiratory rate value is calculated based on the weighted average value.

[0046] Optionally, the fish body pressure level is determined based on the fish's respiratory rate value, including:

[0047] The respiratory rate variation coefficient, peak standard deviation, and energy entropy ratio were determined based on the respiratory rate values ​​of the fish.

[0048] A feature matrix is ​​constructed based on the respiratory rate variation coefficient, peak standard deviation, and energy entropy ratio.

[0049] The feature matrix was weighted using a three-level pressure label and combined with the SHAP value interpretability analysis method to determine the degree of incoordination between the fish mouth and gill movements;

[0050] The fish body pressure level is determined based on the degree of incoordination.

[0051] On the other hand, the present invention also provides a fish respiratory monitoring system, the system including a processor for performing any of the methods described above.

[0052] Through the above technical solution, the embodiments of the present invention provide a method and system for monitoring fish respiration. This method and system extract the motion bounding boxes of the fish mouth and gills using a hybrid cascaded network model. Then, based on these motion bounding boxes, it determines the time-series motion signals of the fish mouth and gills. Finally, it determines the fish's respiratory frequency value based on these time-series motion signals. Compared with existing technologies, the method and system provided by the embodiments of the present invention, through the combination of a hybrid cascaded network model and time-series motion signals, achieve accurate monitoring of fish respiration under the influence of water ripples.

[0053] Other features and advantages of the embodiments of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0054] The accompanying drawings are provided to further illustrate embodiments of the present invention and form part of the specification. They are used together with the following detailed description to explain the embodiments of the present invention, but do not constitute a limitation thereof. In the drawings:

[0055] Figure 1 This is a flowchart of a fish respiration monitoring method according to an embodiment of the present invention;

[0056] Figure 2 This is an example diagram of a hybrid cascaded network model according to one embodiment of the present invention;

[0057] Figure 3 This is a flowchart of a method for training and acquiring a hybrid interconnected network model according to an embodiment of the present invention;

[0058] Figure 4 This is a flowchart of a method for acquiring time-series motion signals according to an embodiment of the present invention;

[0059] Figure 5 This is a flowchart of a method for determining the respiratory rate of a fish according to an embodiment of the present invention;

[0060] Figure 6 This is a flowchart of a method for determining the pressure level of a fish according to an embodiment of the present invention. Detailed Implementation

[0061] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the scope of the present invention.

[0062] In the embodiments of this application, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, they do not mean that the applicant has used or necessarily used the solution.

[0063] like Figure 1 The diagram shown is a flowchart of a fish respiration monitoring method according to an embodiment of the present invention. Figure 1 In this method, the steps may include:

[0064] In step S10, fish video data is acquired;

[0065] In step S11, an image of the fish body video data is captured;

[0066] In step S12, a hybrid cascaded network model is used to perform real-time inference on each frame of video to obtain the motion bounding box of the fish mouth and gills in each frame.

[0067] In step S13, spatiotemporal feature extraction is performed on the motion bounding box to obtain the time-series motion signals of the fish mouth and gills;

[0068] In step S14, the fish's respiratory frequency value is determined based on the time-series motion signal;

[0069] In step S15, the fish pressure level is determined based on the fish's respiratory rate.

[0070] In such Figure 1In the method shown, step S10 can be used to acquire fish video data. The specific method for acquiring this fish video data can be of various forms known to those skilled in the art, such as setting up an image acquisition device underwater or nearby. Further, to improve the acquisition quality of the fish video data, step S10 can be achieved by using a high frame rate (greater than or equal to 120Hz) underwater camera, a wide-angle lens, and an LED cold light source supplementary lighting system to acquire the fish video data in real time. Even further, considering the influence of water ripples, bubbles, etc., step S10 can also first apply Gaussian filtering to the acquired raw video stream to perform low-frequency filtering, thereby eliminating low-frequency noise caused by water ripples in the fish video data. In one example of the invention, the Gaussian filtering operation can be performed using a 5*5 convolution kernel with a σ value of 1.5. Then, for the raw video stream after low-frequency filtering, step S10 can also use median filtering to remove high-frequency salt-and-pepper noise caused by bubbles and suspended particles in the water. In one example of the present invention, the sliding window size of the median filter can be set to 7*7. Finally, to address the color filter problem caused by underwater light absorption, step S10 can also involve adjusting the processed original video stream using a multispectral channel equalization method and enhancing the adjusted original video stream using a contrast-limited adaptive histogram equalization method, thereby obtaining the fish video data.

[0071] Step S11 can be used to capture an image from the fish video data. Specifically, in one example of the present invention, step S11 can be, for example, by presetting a time period and capturing the fish video data at at least one time point in each time period to obtain the captured image.

[0072] Step S12 can be used to perform real-time inference on each frame of video using a hybrid cascaded network model to obtain the motion bounding box of the fish mouth and gills movement in each frame. The specific structure of this hybrid cascaded network model can be of various forms known to those skilled in the art. In one example of the present invention, the hybrid cascaded network model can include, for example... Figure 2 The structure shown is described. Figure 2In this example, the hybrid cascaded network model may include a first convolutional layer (Conv0), a second convolutional layer (Conv_stage1), a third convolutional layer (Conv_stage2), a fourth convolutional layer (Conv_stage3), a fifth convolutional layer (Conv_stage3), a PPM unit, an FPN unit, a feature fusion unit (FUSE), and a feature refinement unit (Refined Feature) connected sequentially. The specific parameters of the first convolutional layer (Conv0), second convolutional layer (Conv_stage1), third convolutional layer (Conv_stage2), fourth convolutional layer (Conv_stage3), fifth convolutional layer (Conv_stage3), PPM unit, FPN unit, feature fusion unit (FUSE), and feature refinement unit (Refined Feature) in this example may be as shown in Table 1 below.

[0073] Table 1

[0074]

[0075] In Figure 2 In this configuration, the first convolutional layer (Conv0), the second convolutional layer (Conv_stage1), the third convolutional layer (Conv_stage2), the fourth convolutional layer (Conv_stage3), and the fifth convolutional layer (Conv_stage3) are connected sequentially. The PPM unit is connected to the output of the fifth convolutional layer (Conv_stage3) and is also connected to the start input of the FPN unit. The FPN unit includes a first upsampling layer C1, a second upsampling layer C2, a third upsampling layer C3, and a fourth upsampling layer C4 connected sequentially. Specifically, the first upsampling layer C1 receives the output of the PPM unit, the second upsampling unit C2 receives the outputs of the fourth convolutional layer and the first upsampling unit C1, the third upsampling unit C3 receives the outputs of the third convolutional layer and the second upsampling unit C2, and the fourth upsampling unit C4 receives the outputs of the second convolutional layer and the third upsampling unit C3. The feature fusion unit is used to fuse the outputs of the first upsampling layer C1, the second upsampling layer C2, the third upsampling layer C3, and the fourth upsampling layer C4. The feature refinement unit is used to refine the feature map output by the feature fusion unit to obtain a feature map with motion bounding boxes.

[0076] The specific method for training and obtaining this hybrid cascaded network model can take many forms known to those skilled in the art. In one example of the present invention, considering the characteristics of fish video data images, the method for training and obtaining this hybrid cascaded network model may include, for example... Figure 3 The steps shown are described in this. Figure 3The method for training and obtaining the hybrid network model may include the following steps:

[0077] In step S20, the fish video is split frame by frame to obtain fish images;

[0078] In step S21, the fish head region in the fish image is labeled;

[0079] In step S22, the labeled fish image is processed by at least one of random rotation, dynamic scaling, fogging simulation, and motion blur. In one example of the present invention, step S22 may employ at least one of random rotation, dynamic scaling, fogging simulation, and motion blur, thereby increasing the data volume of the processed fish image by 10 times.

[0080] In step S23, the processed fish images are combined into a training dataset;

[0081] In step S24, the ConvNeXt network structure is used as the backbone network to construct the initial model. A schematic diagram of the constructed initial model is shown below. Figure 2 As shown;

[0082] In step S25, the loss function of the initial model is determined according to formulas (1) to (3):

[0083] FL(p)=-α(1-p) γ log(p), (1)

[0084] VCE(p)=-log(p), (2)

[0085] Loss(p)=FL(p)+BCE(p), (3)

[0086] Where α is the balance factor, p is the initial predicted probability of the model, γ is the focus factor, and Loss(p) is the loss function. In one example of this invention, to adjust the influence between positive and negative samples, the balance factor α can be set to 0.8, while to adjust the weights between easy and difficult samples, the initial value of the focus factor γ can be set to 2.

[0087] In step S26, the initial model is trained using the dataset to obtain the hybrid cascaded network model.

[0088] Step S13 can be used to perform spatiotemporal feature extraction on the motion bounding box to obtain the time-series motion signal of the fish mouth and gills. Specifically, in one example of the present invention, the method for obtaining the time-series motion signal may include, for example... Figure 4 The steps shown are described in this. Figure 4The method for obtaining the time-series motion signal may include the following steps:

[0089] In step S30, a multi-scale feature pyramid is used to extract the spatiotemporal motion features between the motion bounding boxes of every two frames.

[0090] In step S31, an iterative update mechanism is used to calculate the optical flow field components based on the spatiotemporal motion characteristics. Specifically, in this example, step S31 may involve using a median filter to eliminate anomalous optical flow components in the optical flow field.

[0091] In step S32, the motion intensity characterization is calculated based on the optical flow field components. Specifically, step S32 can be performed using formula (4) to calculate the motion intensity characterization:

[0092]

[0093] in, For the motion intensity characterization of the i-th sampling point, u i v i denoted as different sub-components in the optical flow field component of the i-th sampling point, and λ is the statistical quantile.

[0094] In step S33, a time-series signal is generated based on the motion intensity characterization at a preset sampling rate. In one example of the present invention, the value of the sampling rate may be, for example, set to 120Hz.

[0095] In step S34, a time alignment mechanism is used to compensate for the phase deviation inference of the time series signal using a model (≤30ms).

[0096] In step S35, the time series signal is synchronized based on the result of phase deviation inference to obtain the time series motion signal.

[0097] Step S14 can be used to determine the fish's respiratory frequency value based on the time-series motion signal. Specifically, in one example of the present invention, step S14 may include, for example... Figure 5 The method shown in the diagram. Figure 5 In this context, the method for determining the respiratory rate of a fish may include the following steps:

[0098] In step S40, a fourth-order Butterworth low-pass filter and a Symlet wavelet function (Sym5) are used sequentially to preprocess the time-series motion signal. Specifically, in this example, the cutoff frequency of the fourth-order Butterworth low-pass filter can be 2Hz. The Symlet wavelet function can eliminate baseline drift caused by water turbulence by using soft thresholding with 5-level decomposition.

[0099] In step S41, a sliding window dynamic peak detection algorithm is used to determine respiratory cycle feature points based on the preprocessed time-series motion signal. The window length of this algorithm can be set to 10 seconds, and the step size can be 1 second.

[0100] In step S42, a weighted average of the intervals between adjacent peaks is determined based on the respiratory cycle characteristic points. The specific calculation method for this weighted average can be of various forms known to those skilled in the art. In one example of the present invention, when calculating this weighted average, the weight of the gill cover signal can be set to 0.6, and the weight of the fish mouth signal can be set to 0.4.

[0101] In step S43, the fish's respiratory rate is calculated based on the weighted average value.

[0102] Step S15 can be used to determine the fish's body pressure level based on the fish's respiratory frequency value. The specific method for determining the fish's body pressure level can be of various forms known to those skilled in the art, such as using a preset threshold to classify the fish's respiratory frequency. In one example of the present invention, considering that the present invention uses fish video data, and the fish respiratory frequency obtained based on this fish video data is a continuous data point, therefore, in one example of the present invention, it can be done using methods such as... Figure 6 The method shown is used to determine the fish's body pressure level. Specifically, in this... Figure 6 In determining the fish's body pressure level, the method may include the following steps:

[0103] In step S50, the respiratory rate variation coefficient, peak period standard deviation, and energy entropy ratio are determined based on the fish's respiratory rate value. The energy entropy ratio can be calculated based on the 0.1-0.5Hz (low-frequency stress response) and 0.5-2Hz (high-frequency respiratory rhythm) subbands.

[0104] In step S51, a feature matrix is ​​constructed based on the respiratory rate variation coefficient, peak period standard deviation, and energy entropy ratio.

[0105] In step S52, a three-level pressure label weighted feature matrix is ​​used, combined with the SHAP value interpretability analysis method to determine the degree of incoordination between the fish mouth and gill movements. Specifically, in one example of the present invention, step S52 may involve first inputting the weighted feature matrix into a trained XGBoost classifier, and then determining the degree of incoordination based on the SHAP value output by the XGBoost classifier.

[0106] In step S53, the fish body pressure level is determined based on the degree of incoordination.

[0107] On the other hand, the present invention also provides a fish respiratory monitoring system, the system including a processor for performing any of the methods described above.

[0108] Through the above technical solution, the embodiments of the present invention provide a method and system for monitoring fish respiration. This method and system extract the motion bounding boxes of the fish mouth and gills using a hybrid cascaded network model. Then, based on these motion bounding boxes, it determines the time-series motion signals of the fish mouth and gills. Finally, it determines the fish's respiratory frequency value based on these time-series motion signals. Compared with existing technologies, the method and system provided by the embodiments of the present invention, through the combination of a hybrid cascaded network model and time-series motion signals, achieve accurate monitoring of fish respiration under the influence of water ripples.

[0109] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0110] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0111] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0112] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0113] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0114] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0115] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0116] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0117] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for monitoring fish respiration, characterized in that, include: Acquire fish video data; Extract an image from the fish's video data; A hybrid cascaded network model is used to perform real-time inference on each frame of video to obtain the motion bounding box of the fish mouth and gills in each frame. Spatiotemporal feature extraction is performed on the motion bounding box to obtain the time-series motion signals of the fish mouth and gills; The fish's respiratory rate value is determined based on the time-series motion signal; The fish's body pressure level is determined based on the fish's respiratory rate value. The fish's body pressure level is determined based on the fish's respiratory rate value, including: The respiratory rate variation coefficient, peak standard deviation, and energy entropy ratio were determined based on the respiratory rate values ​​of the fish. A feature matrix is ​​constructed based on the respiratory rate variation coefficient, peak standard deviation, and energy entropy ratio. The feature matrix was weighted using a three-level pressure label and combined with the SHAP value interpretability analysis method to determine the degree of incoordination between the fish mouth and gill movements; The fish body pressure level is determined based on the degree of incoordination.

2. The method according to claim 1, characterized in that, Acquire fish video data, including: Gaussian filtering is used to perform low-frequency filtering on the acquired raw video stream; The median filtering method is used to process the original video stream after low-frequency filtering; The original video stream after processing is adjusted using a multispectral channel equalization method; The contrast-limited adaptive histogram equalization method is used to enhance the adjusted original video stream.

3. The method according to claim 1, characterized in that, Also includes: The fish video is split frame by frame to obtain the fish image; The head region of the fish in the fish image is labeled; The labeled fish images are subjected to at least one of the following processing methods: random rotation, dynamic scaling, fogging simulation, and motion blur. The processed fish images are combined into a training dataset. The initial model was constructed using the ConvNeXt network structure as the backbone network. The loss function of the initial model is determined according to formulas (1) to (3): ，（1） ，（2） ，（3） in, As a balance factor, The predicted probabilities of the initial model. As the focal factor, The loss function is... The initial model is trained using the dataset to obtain the hybrid cascaded network model.

4. The method according to claim 1, characterized in that, Perform spatiotemporal feature extraction on the motion bounding box to obtain the time-series motion signals of the fish mouth and gills, including: A multi-scale feature pyramid is used to extract the spatiotemporal motion features between motion bounding boxes of every two frames; An iterative update mechanism is used to calculate the optical flow field components based on the aforementioned spatiotemporal motion characteristics; The motion intensity characterization is calculated based on the optical flow field components. A time-series signal is generated based on the motion intensity characterization at a preset sampling rate; A time alignment mechanism compensation model is used to perform phase deviation inference on the time series signal; The time series signal is synchronized based on the phase deviation inference result to obtain the time series motion signal.

5. The method according to claim 4, characterized in that, The optical flow field components are calculated based on the spatiotemporal motion characteristics using an iterative update mechanism, including: A median filter is used to eliminate abnormal optical flow components in the optical flow field.

6. The method according to claim 4, characterized in that, The motion intensity characterization is calculated based on the optical flow field components, including: The intensity of the exercise is calculated using formula (4): ，（4） in, For the first The motion intensity characterization of each sampling point , The first Different sub-components of the optical flow field at each sampling point For statistical quantiles.

7. The method according to claim 1, characterized in that, Determining the fish's respiratory rate based on the time-series motion signal includes: The time-series motion signal is preprocessed sequentially using a fourth-order Butterworth low-pass filter and a Symlet wavelet function. A sliding window dynamic peak detection algorithm is used to determine respiratory cycle feature points based on the preprocessed time-series motion signal. The weighted average of the intervals between adjacent peaks is determined based on the respiratory cycle characteristic points; The fish's respiratory rate value is calculated based on the weighted average value.

8. A fish respiratory monitoring system, characterized in that, The system includes a processor for performing the method as described in any one of claims 1 to 7.

Images