Intelligent sturgeon health monitoring method and system based on multi-source data fusion

Abstract

The invention provides an intelligent sturgeon health monitoring method and system based on multi-source data fusion, and the method comprises the steps: firstly, synchronously collecting underwater image data and infrared thermal image data of a sturgeon culture pond in real time, and carrying out the preprocessing of the data; the method comprises the steps that sturgeon behavior features and fish body temperature features are extracted, then fish body positioning is carried out, behavior feature extraction and temperature feature extraction are carried out respectively, feature fusion is carried out on the sturgeon behavior features and the fish body temperature features, finally, the sturgeon fusion features are input into a pre-trained anomaly detection model, and an anomaly detection result is obtained; according to the system, comprehensive, real-time and intelligent monitoring of the sturgeon culture process is achieved by integrating various sensing technologies and an advanced data analysis algorithm, abnormal conditions can be found and processed in time, and scientific basis and optimization suggestions can be provided for culture management through long-term data accumulation and analysis; meanwhile, the sturgeon breeding efficiency and the product quality are also remarkably improved, and the disease risk and the economic loss are reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of aquaculture, and more specifically, to a sturgeon health intelligent monitoring method and system based on multi-source data fusion. Background Art

[0002] Sturgeon is an aquaculture species with high economic value, and its caviar is highly favored in the international market. However, there are many challenges in the process of sturgeon farming, the most prominent of which is the difficulty in timely detection and treatment of fish diseases. Traditional sturgeon farming monitoring methods mainly rely on manual inspections, which have the following disadvantages:

[0003] 1) Low monitoring frequency: Manual inspections are usually conducted a limited number of times per day, making it difficult to achieve 24 / 7 monitoring;

[0004] 2) Limited observation range: The water depth of the aquaculture pond is usually about 1.5 meters, and it is difficult for the human eye to fully observe the underwater situation;

[0005] 3) High subjectivity: Different staff members may have different experiences and judgment criteria, leading to inconsistent monitoring results;

[0006] 4) Delayed response: When fish abnormalities are visible to the naked eye, the disease may have already developed to a more serious stage;

[0007] 5) Difficulty in data recording and analysis: Manual monitoring makes it difficult to form systematic data accumulation and analysis.

[0008] At present, some intelligent breeding systems have emerged, which have improved the monitoring capabilities of aquaculture in some aspects. For example, an aquaculture monitoring system is disclosed in an existing patent document, including a pool water collector set in the pool to be monitored, a water quality collection device set in the pool water collector, a sensor module set in the water quality collection device, the sensor module is used to monitor the water temperature, the oxygen content of the water, the pH value of the water, and the dissolved oxygen content of the water. The sensor is connected to the microprocessor through the control module. The microprocessor is set on the monitoring base station. At the same time, the monitoring base station is connected with an output device display, a command input device keyboard, a data storage device and a power supply. The microprocessor receives and transmits the command of the control module. At the same time, the microprocessor is connected with a sewage discharge device, a water pump, an oxygen supply device, a water temperature adjustment device, a camera, an alarm module, a drain valve, and a salinity adjustment device. However, the system collects breeding environment data through water quality sensors, cameras and other equipment, but mainly focuses on water quality monitoring, and has limited ability to identify abnormal fish behavior.

[0009] In addition, the existing intelligent breeding system still has the following shortcomings:

[0010] 1) Single data source: Most systems rely on only a single type of sensor data, which is difficult to fully reflect the health status of fish;

[0011] 2) Insufficient intelligent analysis capabilities: lack of fusion analysis of multi-source data, unable to fully utilize the complementarity of various types of data;

[0012] 3) Lack of specificity: Most of them are general aquaculture monitoring systems, which are not optimized for the characteristics of sturgeons and have low monitoring accuracy;

[0013] 4) Inconvenient human-computer interaction: It is difficult for managers to obtain monitoring information and perform remote operations anytime and anywhere. Summary of the invention

[0014] In order to overcome the defects of the above-mentioned prior art that manual monitoring is inefficient and has low real-time performance and is highly subjective, and that the intelligent monitoring system has a single data source, low monitoring accuracy and is not comprehensive, the present invention provides a sturgeon health intelligent monitoring method and system based on multi-source data fusion, which can comprehensively, intelligently and in real time monitor the health status of sturgeons, so as to improve breeding efficiency and reduce economic losses.

[0015] In order to solve the above technical problems, the technical solution of the present invention is as follows:

[0016] A sturgeon health intelligent monitoring method based on multi-source data fusion includes the following steps:

[0017] S1: Real-time synchronous acquisition of underwater image data and infrared thermal image data of sturgeon breeding ponds, and pre-processing them respectively;

[0018] S2: inputting the pre-processed underwater image data into a pre-trained fish body detection model, locating all sturgeons in each image, obtaining the sturgeon position information, and searching for sturgeon fish bodies in the pre-processed infrared thermal image data according to the sturgeon position information;

[0019] S3: Performing behavior analysis on the sturgeon according to the sturgeon position information to obtain the behavior characteristics of the sturgeon; the behavior analysis includes: swimming trajectory analysis, swimming speed analysis, posture analysis and group behavior analysis;

[0020] Extracting the body temperature characteristics of each sturgeon from the preprocessed infrared thermal imaging data;

[0021] S4: fusing the sturgeon behavior characteristics and the fish body temperature characteristics to obtain sturgeon fusion characteristics;

[0022] S5: Input the sturgeon fusion features into a pre-trained anomaly detection model to obtain anomaly detection results, and complete intelligent monitoring of sturgeon health.

[0023] Preferably, in step S1, the following operations are performed on the underwater image data to obtain pre-processed underwater image data:

[0024] Image denoising: Use Gaussian filtering or median filtering to remove noise caused by the underwater environment;

[0025] Contrast enhancement: The contrast-limited adaptive histogram equalization algorithm is used to improve the clarity of underwater image data;

[0026] Color correction: Select a reference white point from the underwater image data, calculate the average value of the reference white point on the three color channels of red, green and blue, and compare it with the preset standard values ​​of the three color channels to obtain the gain coefficient of each color channel; use the gain coefficients of the three color channels of red, green and blue to adjust the white balance of the underwater image data;

[0027] Perform the following operations on the infrared thermal imaging data to obtain pre-processed infrared thermal imaging data:

[0028] Image segmentation: The infrared thermal image data is converted into a grayscale image, and the segmentation threshold of the water surface and the fish body area is calculated using the OTUS algorithm. Figure 2 The water surface and fish body area are separated by quantization.

[0029] Preferably, in step S2, the fish body detection model is specifically: a YOLOv8 target detection model, and the sturgeon position information includes: the coordinates of the center point of the sturgeon body and the contour points of the sturgeon body.

[0030] Preferably, in step S2, searching for the sturgeon body in the preprocessed infrared thermal imaging data according to the sturgeon position information comprises the following steps:

[0031] Affine transformation is used to establish the coordinate system mapping relationship between underwater image data and infrared thermal imaging data, which is expressed as:

[0032]

[0033] Among them, (x i ,y i ) is the coordinate of the i-th pixel point in the underwater image data coordinate system, (x′ i ,y′ i ) is the corresponding pixel coordinate in the infrared thermal image data coordinate system, (x i ,y i ) and (x′ i ,y′ i ) constitute a set of pixel pairs; is the linear transformation matrix, is the translation matrix;

[0034] Use n sets of pixel pairs to construct the following linear equations:

[0035]

[0036] Wherein, n is a positive integer;

[0037] The constructed linear equations are solved using the least squares method to obtain the calculation results of the linear transformation matrix and the translation matrix;

[0038] According to the calculation result, the sturgeon position information is mapped to the preprocessed infrared thermal imaging data, and a plurality of mapping centers representing the sturgeon positions are formed in the infrared thermal imaging data. A search window is set with each mapping coordinate as the center, and a connected area with a temperature higher than a preset value is searched in the search window to obtain the corresponding sturgeon body.

[0039] Preferably, in step S3, extracting the body temperature characteristics of each sturgeon includes:

[0040] In the pre-processed infrared thermal imaging data, for each located sturgeon body area, the temperature values ​​of all pixels in the area are extracted, the temperature mean of all pixels in the area is calculated as the average body temperature of the corresponding sturgeon, and the average temperature and standard deviation of all sturgeons in the school of fish are further calculated;

[0041] Calculate the temperature change rate based on the average temperature of the fish body at each moment;

[0042] The fish body average temperature and the temperature change rate are stored together as the fish body temperature characteristics of the sturgeon.

[0043] Preferably, in step S3, the swimming trajectory analysis includes:

[0044] Based on the position of the fish center at the previous moment, the Kalman filter algorithm is used to predict and update the position of the fish center at the next moment in real time, obtain the movement trajectory of the fish center over a period of time, and further obtain the length, curvature and movement direction change frequency of the movement trajectory. The length, curvature and movement direction change frequency of the movement trajectory are saved together as the swimming trajectory features to complete the swimming trajectory analysis;

[0045] The swimming speed analysis includes:

[0046] The displacement of the fish body per unit time is calculated based on the coordinates of the center point of the fish body at several moments, and the instantaneous speed of the sturgeon is further calculated. The average speed over a period of time is calculated based on the instantaneous speed of the sturgeon, and the average speed is saved as the swimming speed feature to complete the swimming speed analysis;

[0047] The posture analysis includes:

[0048] The principal component analysis method is used to calculate the principal axis direction of the fish body contour, specifically:

[0049] Calculate the covariance matrix Cov of the fish body contour points:

[0050]

[0051] Among them, (x j ,y j ) is the coordinate of the jth pixel point in the fish body contour point; x and y are the mean value of the horizontal coordinate and the mean value of the vertical coordinate of the fish body contour point respectively;

[0052] Construct the characteristic equation: Cov·v=λ·v, where λ and v are the eigenvalue and eigenvector of the covariance matrix Cov respectively;

[0053] Solving the characteristic equation to obtain a first eigenvalue λ1 and a first eigenvector v1 corresponding thereto, and a second eigenvalue λ2 and a second eigenvector v2 corresponding thereto;

[0054] The eigenvector corresponding to the maximum eigenvalue is used as the main axis direction of the fish body contour, the main axis direction is used to calculate the inclination angle of the sturgeon swimming, and the inclination angle is saved as a posture feature to complete the posture analysis;

[0055] The group behavior analysis includes:

[0056] Movement consistency analysis: The instantaneous speed of each sturgeon is used to calculate the consistency index C of the fish school:

[0057]

[0058] Among them, u k is the instantaneous speed of the sturgeon at the kth moment; N is the number of sturgeons in the school of fish; the larger the consistency index C is, the more consistent the movement of the school of fish is;

[0059] Spatial distribution analysis: The spatial distribution of fish schools was evaluated using the nearest neighbor distance method, the Euclidean distance between each sturgeon and its nearest neighbor was calculated, the nearest neighbor distance of each sturgeon was obtained, and the average nearest neighbor distance of the fish school was calculated;

[0060] The nearest neighbor distance of each sturgeon is saved as a group behavior feature to complete the group behavior analysis;

[0061] The swimming trajectory characteristics, swimming speed characteristics, posture characteristics and group behavior characteristics are stored together as the sturgeon behavior characteristics.

[0062] Preferably, in step S4, the sturgeon behavior feature and the fish body temperature feature are fused according to the following formula to obtain the sturgeon fusion feature F fused :

[0063]

[0064] Among them, f1~f s Respectively represent the 1st to sth characteristic index values ​​of sturgeon; f 1,max ~f s,max They represent the maximum values ​​of the 1st to sth characteristic index values ​​of sturgeon respectively; f 1,min ~f s,min represent the minimum values ​​of the 1st to sth characteristic index values ​​of sturgeon; w1~w s They represent the weight factors corresponding to the 1st to sth characteristic index values ​​of sturgeon respectively;

[0065] The characteristic indicators of the sturgeon include: any one of the length of the motion trajectory, the curvature, the frequency of change of the motion direction, the average speed, the tilt angle, the nearest neighbor distance, the average temperature of the fish body and the temperature change rate.

[0066] Preferably, in step S5, the anomaly detection model is specifically a multi-layer LSTM neural network model;

[0067] The multi-layer LSTM neural network model includes: an input layer, a plurality of sequentially connected LSTM layers, a fully connected layer, and an output layer;

[0068] A sliding window is set, and a time series sample is generated using the sturgeon fusion features obtained in the historical time period, and the samples are divided into a training set, a validation set, and a test set. The multi-layer LSTM neural network model is supervised for training using a binary cross entropy function, and a trained anomaly detection model is obtained after verification and testing;

[0069] The sturgeon fusion features obtained in real time are input into the trained anomaly detection model to obtain anomaly detection results, and anomaly alarm is performed according to the anomaly detection results; the anomaly detection results are the anomaly scores of each feature index of the sturgeon in each time window.

[0070] Preferably, the trained anomaly detection model determines whether each feature indicator is abnormal based on the following rules:

[0071] If the length of the sturgeon's movement trajectory is less than the first preset threshold, the sturgeon's behavior is abnormal;

[0072] If the curvature of the sturgeon's motion trajectory is greater than a second preset threshold, the sturgeon's behavior is abnormal;

[0073] If the frequency of the sturgeon's movement direction changes is greater than a third preset threshold, the sturgeon's behavior is abnormal;

[0074] If the average speed of the sturgeon is less than the fourth preset threshold, the sturgeon behaves abnormally;

[0075] If the sturgeon's tilt angle is greater than a fifth preset threshold, the sturgeon's behavior is abnormal;

[0076] If the absolute value of the difference between the nearest neighbor distance of the sturgeon and the average nearest neighbor distance of the fish school is greater than a sixth preset threshold, the sturgeon behaves abnormally;

[0077] If any of the following conditions occur, the sturgeon temperature is abnormal:

[0078] The average body temperature of the sturgeon exceeds the preset normal temperature range, the absolute value of the difference between the average body temperature of the sturgeon and the average temperature of all sturgeons in the school is greater than two standard deviations, and the temperature change rate is greater than the seventh preset threshold.

[0079] The present invention also provides a sturgeon health intelligent monitoring system based on multi-source data fusion, which uses the above-mentioned sturgeon health intelligent monitoring method based on multi-source data fusion, including:

[0080] Preprocessing unit: used to synchronously collect underwater image data and infrared thermal image data of sturgeon breeding ponds in real time and perform preprocessing respectively;

[0081] A fish body positioning unit is used to input the pre-processed underwater image data into a pre-trained fish body detection model, locate all sturgeons in each image, obtain the sturgeon position information, and search for sturgeons in the pre-processed infrared thermal image data according to the sturgeon position information;

[0082] Feature extraction unit: used to perform behavior analysis on the sturgeon according to the position information of the sturgeon to obtain the behavior characteristics of the sturgeon; the behavior analysis includes: swimming trajectory analysis, swimming speed analysis, posture analysis and group behavior analysis;

[0083] and for extracting the body temperature characteristics of each sturgeon in the pre-processed infrared thermal imaging data;

[0084] Feature fusion unit: used for fusing the sturgeon behavior feature and the fish body temperature feature to obtain the sturgeon fusion feature;

[0085] Anomaly detection unit: used to input the sturgeon fusion features into a pre-trained anomaly detection model, obtain anomaly detection results, and complete intelligent monitoring of sturgeon health.

[0086] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0087] The present invention provides a sturgeon health intelligent monitoring method and system based on multi-source data fusion. First, underwater image data and infrared thermal image data of a sturgeon breeding pond are synchronously collected in real time and preprocessed respectively; then, the preprocessed underwater image data is input into a pre-trained fish body detection model, all sturgeons in each image are located, and the sturgeon position information is obtained; the sturgeon body in the pre-processed infrared thermal image data is searched according to the sturgeon position information, and the body temperature characteristics of each sturgeon are extracted; the sturgeon behavior is analyzed according to the sturgeon position information to obtain the sturgeon behavior characteristics; the behavior analysis includes: swimming trajectory analysis, swimming speed analysis, posture analysis and group behavior analysis; the sturgeon behavior characteristics and the fish body temperature characteristics are feature fused to obtain the sturgeon fusion characteristics; finally, the sturgeon fusion characteristics are input into a pre-trained anomaly detection model to obtain the anomaly detection results, and the intelligent monitoring of sturgeon health is completed;

[0088] The beneficial effects of the present invention are as follows:

[0089] 1) Comprehensive monitoring: Underwater images and infrared thermal images obtained by multiple sensors such as underwater cameras and infrared thermal imaging enable all-round monitoring of sturgeon behavior (swimming trajectory, swimming speed, posture analysis, group behavior) and body temperature, improving the accuracy and comprehensiveness of anomaly detection.

[0090] 2) Strong real-time performance: The present invention can monitor 24 hours a day and detect abnormal situations in a timely manner, greatly shortening the time interval from problem discovery to problem handling.

[0091] 3) High degree of intelligence: Combined with artificial intelligence technologies such as machine learning (target recognition model and anomaly detection model), it can automatically identify complex abnormal patterns based on multi-source data, reduce human judgment errors, and improve detection efficiency and accuracy.

[0092] 4) Data fusion advantage: Through multi-source data fusion technology (time synchronization, spatial alignment and feature fusion of multi-source data), the complementarity of different types of data is fully utilized to improve the reliability and robustness of the system.

[0093] 5) Convenient remote monitoring: Monitoring data and alarm information can be sent to the terminal of the breeding management personnel, which can be viewed anytime and anywhere, thus improving management efficiency.

[0094] 6) Strong scalability: The overall architecture of the present invention is flexible in design and can be easily integrated with other types of sensors and analysis modules to adapt to future technological developments and changes in demand.

[0095] 7) Significant economic benefits: By early detection and treatment of fish health problems, the economic losses caused by diseases are effectively reduced and the breeding benefits are improved.

[0096] 8) Data value mining: The data accumulated over a long period of time in the present invention can be used for in-depth analysis and mining, providing a scientific basis for optimizing breeding strategies and preventing diseases.

[0097] 9) Wide applicability: The basic principle and architecture of the present invention can also be applied to other aquaculture species and has broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0098] Figure 1 This is a flow chart of a method for intelligent monitoring of sturgeon health based on multi-source data fusion provided in Example 1.

[0099] Figure 2 This is a flow chart of a method for intelligent monitoring of sturgeon health based on multi-source data fusion provided in Example 2.

[0100] Figure 3 This is a hardware architecture diagram of the sturgeon health intelligent monitoring method based on multi-source data fusion provided in Example 2.

[0101] Figure 4 This is a functional diagram of the central control platform provided in Example 2. DETAILED DESCRIPTION

[0102] The accompanying drawings are only used for illustrative purposes and are not to be construed as limiting the present invention;

[0103] In order to better illustrate the present embodiment, some parts in the drawings may be omitted, enlarged or reduced, and do not represent the size of the actual product;

[0104] It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0105] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0106] Example 1

[0107] like Figure 1 As shown, this embodiment provides a sturgeon health intelligent monitoring method based on multi-source data fusion, comprising the following steps:

[0108] S1: Real-time synchronous acquisition of underwater image data and infrared thermal image data of sturgeon breeding ponds, and pre-processing them respectively;

[0109] S2: inputting the pre-processed underwater image data into a pre-trained fish body detection model, locating all sturgeons in each image, obtaining the sturgeon position information, and searching for sturgeon fish bodies in the pre-processed infrared thermal image data according to the sturgeon position information;

[0110] S3: Performing behavior analysis on the sturgeon according to the sturgeon position information to obtain the behavior characteristics of the sturgeon; the behavior analysis includes: swimming trajectory analysis, swimming speed analysis, posture analysis and group behavior analysis;

[0111] Extracting the body temperature characteristics of each sturgeon from the preprocessed infrared thermal imaging data;

[0112] S4: fusing the sturgeon behavior characteristics and the fish body temperature characteristics to obtain sturgeon fusion characteristics;

[0113] S5: Input the sturgeon fusion features into a pre-trained anomaly detection model to obtain anomaly detection results, and complete intelligent monitoring of sturgeon health.

[0114] In the specific implementation process, firstly, the underwater image data and infrared thermal image data of the sturgeon breeding pond are collected synchronously in real time and preprocessed respectively; then, the preprocessed underwater image data are input into the pre-trained fish body detection model, all sturgeons in each image are located, and the sturgeon position information is obtained; the sturgeon body in the pre-processed infrared thermal image data is searched according to the sturgeon position information, and the body temperature characteristics of each sturgeon are extracted; the behavior of the sturgeon is analyzed according to the sturgeon position information to obtain the sturgeon behavior characteristics; the behavior analysis includes: swimming trajectory analysis, swimming speed analysis, posture analysis and group behavior analysis; the sturgeon behavior characteristics and fish body temperature characteristics are fused to obtain the sturgeon fusion characteristics; finally, the sturgeon fusion characteristics are input into the pre-trained anomaly detection model to obtain the anomaly detection results, and the intelligent monitoring of sturgeon health is completed;

[0115] This method achieves comprehensive, real-time, and intelligent monitoring of the sturgeon farming process by integrating multiple sensing technologies and advanced data analysis algorithms. It can not only detect and handle abnormal situations in a timely manner, but also provide a scientific basis and optimization suggestions for farming management through long-term data accumulation and analysis.

[0116] Example 2

[0117] This embodiment provides a method for intelligent monitoring of sturgeon health based on multi-source data fusion, comprising the following steps:

[0118] S1: Real-time synchronous acquisition of underwater image data and infrared thermal image data of sturgeon breeding ponds, and pre-processing them respectively;

[0119] S2: inputting the pre-processed underwater image data into a pre-trained fish body detection model, locating all sturgeons in each image, obtaining the sturgeon position information, and searching for sturgeon fish bodies in the pre-processed infrared thermal image data according to the sturgeon position information;

[0120] S3: Performing behavior analysis on the sturgeon according to the sturgeon position information to obtain the behavior characteristics of the sturgeon; the behavior analysis includes: swimming trajectory analysis, swimming speed analysis, posture analysis and group behavior analysis;

[0121] Extracting the body temperature characteristics of each sturgeon from the preprocessed infrared thermal imaging data;

[0122] S4: fusing the sturgeon behavior characteristics and the fish body temperature characteristics to obtain sturgeon fusion characteristics;

[0123] S5: inputting the sturgeon fusion feature into a pre-trained anomaly detection model to obtain anomaly detection results, thereby completing intelligent monitoring of sturgeon health;

[0124] In step S1, the following operations are performed on the underwater image data to obtain pre-processed underwater image data:

[0125] Image denoising: Use Gaussian filtering or median filtering to remove noise caused by the underwater environment;

[0126] Contrast enhancement: The contrast-limited adaptive histogram equalization algorithm is used to improve the clarity of underwater image data;

[0127] Color correction: Select a reference white point from the underwater image data, calculate the average value of the reference white point on the three color channels of red, green and blue, and compare it with the preset standard values ​​of the three color channels to obtain the gain coefficient of each color channel; use the gain coefficients of the three color channels of red, green and blue to adjust the white balance of the underwater image data;

[0128] Perform the following operations on the infrared thermal imaging data to obtain pre-processed infrared thermal imaging data:

[0129] Image segmentation: The infrared thermal image data is converted into a grayscale image, and the segmentation threshold of the water surface and the fish body area is calculated using the OTUS algorithm. Figure 2 Value, separate the water surface and fish body area;

[0130] In the step S2, the fish body detection model is specifically: a YOLOv8 target detection model, and the sturgeon position information includes: the coordinates of the center point of the sturgeon body and the contour points of the sturgeon body;

[0131] In the step S2, searching for the sturgeon body in the pre-processed infrared thermal imaging data according to the sturgeon position information includes the following steps:

[0132] Affine transformation is used to establish the coordinate system mapping relationship between underwater image data and infrared thermal imaging data, which is expressed as:

[0133]

[0134] Among them, (x i ,y i ) is the coordinate of the i-th pixel point in the underwater image data coordinate system, (x′ i ,y′ i ) is the corresponding pixel coordinate in the infrared thermal image data coordinate system, (x i ,y i ) and (x′ i ,y′ i ) constitute a set of pixel pairs; is the linear transformation matrix, is the translation matrix;

[0135] Use n sets of pixel pairs to construct the following linear equations:

[0136]

[0137] Wherein, n is a positive integer;

[0138] The constructed linear equations are solved using the least squares method to obtain the calculation results of the linear transformation matrix and the translation matrix;

[0139] According to the calculation result, the sturgeon position information is mapped to the preprocessed infrared thermal imaging data, a plurality of mapping centers representing the sturgeon position are formed in the infrared thermal imaging data, and a search window is set with each mapping coordinate as the center, and a connected area with a temperature higher than a preset value is searched in the search window to obtain the corresponding sturgeon body;

[0140] In step S3, extracting the body temperature characteristics of each sturgeon includes:

[0141] In the pre-processed infrared thermal imaging data, for each located sturgeon body area, the temperature values ​​of all pixels in the area are extracted, the temperature mean of all pixels in the area is calculated as the average body temperature of the corresponding sturgeon, and the average temperature and standard deviation of all sturgeons in the school of fish are further calculated;

[0142] Calculate the temperature change rate based on the average temperature of the fish body at each moment;

[0143] The fish body average temperature and the temperature change rate are stored together as the fish body temperature characteristics of the sturgeon;

[0144] In step S3, the swimming trajectory analysis includes:

[0145] Based on the position of the fish center at the previous moment, the Kalman filter algorithm is used to predict and update the position of the fish center at the next moment in real time, obtain the movement trajectory of the fish center over a period of time, and further obtain the length, curvature and movement direction change frequency of the movement trajectory. The length, curvature and movement direction change frequency of the movement trajectory are saved together as the swimming trajectory features to complete the swimming trajectory analysis;

[0146] The swimming speed analysis includes:

[0147] The displacement of the fish body per unit time is calculated based on the coordinates of the center point of the fish body at several moments, and the instantaneous speed of the sturgeon is further calculated. The average speed over a period of time is calculated based on the instantaneous speed of the sturgeon, and the average speed is saved as the swimming speed feature to complete the swimming speed analysis;

[0148] The posture analysis includes:

[0149] The principal component analysis method is used to calculate the principal axis direction of the fish body contour, specifically:

[0150] Calculate the covariance matrix Cov of the fish body contour points:

[0151]

[0152] Among them, (x j ,y j ) is the coordinate of the jth pixel point in the fish body contour point; and are the mean values ​​of the horizontal and vertical coordinates of the fish body contour points respectively;

[0153] Construct the characteristic equation: Cov·v=λ·v, where λ and v are the eigenvalue and eigenvector of the covariance matrix Cov respectively;

[0154] Solving the characteristic equation to obtain a first eigenvalue λ1 and a first eigenvector v1 corresponding thereto, and a second eigenvalue λ2 and a second eigenvector v2 corresponding thereto;

[0155] The eigenvector corresponding to the maximum eigenvalue is used as the main axis direction of the fish body contour, the main axis direction is used to calculate the inclination angle of the sturgeon swimming, and the inclination angle is saved as a posture feature to complete the posture analysis;

[0156] The group behavior analysis includes:

[0157] Movement consistency analysis: The instantaneous speed of each sturgeon is used to calculate the consistency index C of the fish school:

[0158]

[0159] Among them, u kis the instantaneous speed of the sturgeon at the kth moment; N is the number of sturgeons in the school of fish; the larger the consistency index C is, the more consistent the movement of the school of fish is;

[0160] Spatial distribution analysis: The spatial distribution of fish schools was evaluated using the nearest neighbor distance method, the Euclidean distance between each sturgeon and its nearest neighbor was calculated, the nearest neighbor distance of each sturgeon was obtained, and the average nearest neighbor distance of the fish school was calculated;

[0161] The nearest neighbor distance of each sturgeon is saved as a group behavior feature to complete the group behavior analysis;

[0162] The swimming trajectory characteristics, swimming speed characteristics, posture characteristics and group behavior characteristics are collectively saved as the sturgeon behavior characteristics;

[0163] In step S4, the sturgeon behavior characteristics and the fish body temperature characteristics are fused according to the following formula to obtain the sturgeon fusion feature F fused :

[0164]

[0165] Among them, f1~f s Respectively represent the 1st to sth characteristic index values ​​of sturgeon; f 1,max ~f s,max They represent the maximum values ​​of the 1st to sth characteristic index values ​​of sturgeon respectively; f 1,min ~f s,min represent the minimum values ​​of the 1st to sth characteristic index values ​​of sturgeon; w1~w s They represent the weight factors corresponding to the 1st to sth characteristic index values ​​of sturgeon respectively;

[0166] The characteristic indicators of the sturgeon include: any one of the length of the motion trajectory, curvature, frequency of motion direction change, average speed, tilt angle, nearest neighbor distance, average temperature of the fish body and temperature change rate;

[0167] In step S5, the anomaly detection model is specifically a multi-layer LSTM neural network model;

[0168] The multi-layer LSTM neural network model includes: an input layer, a plurality of sequentially connected LSTM layers, a fully connected layer, and an output layer;

[0169] A sliding window is set, and a time series sample is generated using the sturgeon fusion features obtained in the historical time period, and the samples are divided into a training set, a validation set, and a test set. The multi-layer LSTM neural network model is supervised for training using a binary cross entropy function, and a trained anomaly detection model is obtained after verification and testing;

[0170] Input the sturgeon fusion features obtained in real time into the trained anomaly detection model to obtain anomaly detection results, and issue an anomaly alarm based on the anomaly detection results; the anomaly detection results are the anomaly scores of each feature index of the sturgeon in each time window;

[0171] The trained anomaly detection model determines whether each feature indicator is abnormal based on the following rules:

[0172] If the length of the sturgeon's movement trajectory is less than the first preset threshold, the sturgeon's behavior is abnormal;

[0173] If the curvature of the sturgeon's motion trajectory is greater than a second preset threshold, the sturgeon's behavior is abnormal;

[0174] If the frequency of the sturgeon's movement direction changes is greater than a third preset threshold, the sturgeon's behavior is abnormal;

[0175] If the average speed of the sturgeon is less than the fourth preset threshold, the sturgeon behaves abnormally;

[0176] If the sturgeon's tilt angle is greater than a fifth preset threshold, the sturgeon's behavior is abnormal;

[0177] If the absolute value of the difference between the nearest neighbor distance of the sturgeon and the average nearest neighbor distance of the fish school is greater than a sixth preset threshold, the sturgeon behaves abnormally;

[0178] If any of the following conditions occur, the sturgeon temperature is abnormal:

[0179] The average body temperature of the sturgeon exceeds the preset normal temperature range, the absolute value of the difference between the average body temperature of the sturgeon and the average temperature of all sturgeons in the school is greater than two standard deviations, and the temperature change rate is greater than the seventh preset threshold.

[0180] In the specific implementation process, Figure 2 As shown in the flowchart, this method is mainly divided into four processes: data collection, data processing, analysis and decision-making, and result output, which will be described in detail below;

[0181] Firstly, underwater image data and infrared thermal image data of sturgeon breeding ponds are collected synchronously in real time and preprocessed respectively;

[0182] like Figure 3 As shown, the hardware architecture of this method is based on underwater cameras, infrared thermal imagers, central control platforms and intelligent terminals;

[0183] In this embodiment, 3 to 5 high-definition cameras are installed in each breeding pond to cover different angles and depths and simultaneously obtain underwater image data; the high-definition underwater camera adopts an industrial-grade camera with a waterproof grade of IP68, a resolution of not less than 1080p, and a frame rate of not less than 30fps; the camera is connected to a waterproof junction box at the edge of the pool through a waterproof cable, and then the data is transmitted to the central control platform through an optical fiber network; 1 to 2 infrared thermal imagers are installed above each breeding pond to collect infrared thermal imaging data, and the infrared thermal imager is fixed by an adjustable bracket to cover the entire water surface; the infrared thermal imager adopts a non-cooled thermal imager with a resolution of not less than 640×480, and a temperature resolution better than 0.05°C; the thermal imager is connected to a data collector at the edge of the pool through a waterproof cable, and then the data is transmitted to the central control platform through Ethernet;

[0184] like Figure 4 As shown, the central control platform provides functions such as data reception, data storage, data fusion, intelligent analysis and alarm generation, and sends the generated alarm information to intelligent terminals (mobile phones, computers, etc.) for management personnel to view, so as to realize real-time monitoring; in this embodiment, the central control platform uses TCP / IP protocol to receive data from underwater cameras and infrared thermal imagers, and realizes data buffering mechanism to cope with network fluctuations and data emergencies; uses time series database InfluxDB to store sensor data to optimize the storage and query efficiency of time series data, and uses relational database PostgreSQL to store structured data such as fish body information and equipment configuration, realizes data sharding and backup strategy, and ensures data reliability and scalability; in addition, the received raw data is preliminarily cleaned and compressed to reduce space occupancy;

[0185] In order to ensure synchronous data acquisition, the central control platform uses the NTP (Network Time Protocol) protocol to ensure the consistency of timestamps of different sensors and implement a time window mechanism to align data from different sensors in time.

[0186] Preprocessing of underwater image data includes:

[0187] Image denoising: Use Gaussian filtering or median filtering to remove noise caused by the underwater environment;

[0188] For Gaussian filtering, the two-dimensional convolution operation of Gaussian filtering can be expressed as:

[0189]

[0190] Where (x, y) represents the pixel coordinates, σ is the standard deviation of the Gaussian distribution;

[0191] For median filtering, a 3×3 or 5×5 sliding window can be used to replace the central pixel with the median of all pixels in the window, effectively removing salt and pepper noise;

[0192] Contrast enhancement: The contrast limited adaptive histogram equalization (CLAHE) algorithm is used to improve the clarity of underwater image data. The CLAHE algorithm divides the image into several small blocks, calculates the grayscale histogram of each small block, and performs cropping, and redistributes the pixel values ​​exceeding the preset threshold to other gray levels to limit the contrast enhancement amplitude. Then, each small block is histogram equalized, and the boundaries between adjacent small blocks are bilinearly interpolated to eliminate the block effect and obtain a smooth enhanced image.

[0193] Color correction: Select a reference white point from the underwater image data, calculate the average value of the reference white point on the three color channels of red, green and blue (R, G and B), and compare them with the preset standard values ​​of the three color channels to obtain the gain coefficients of each color channel; use the gain coefficients of the three color channels to adjust the white balance of the underwater image data, so as to restore the true color of the image;

[0194] For infrared thermal imaging data preprocessing, including:

[0195] Image segmentation: The infrared thermal image data is converted into a grayscale image, and the segmentation threshold of the water surface and the fish body area is calculated using the OTUS algorithm. The OTUS algorithm is an automatic threshold selection method based on the image histogram, which can maximize the inter-class variance between the target and the background, thereby achieving the best image segmentation effect; the OTUS algorithm steps are as follows:

[0196] 1) Calculate the image histogram: Count the number of pixels that appear at each gray level in the image to obtain the grayscale histogram of the image;

[0197] 2) Calculate intra-class variance and inter-class variance: For each possible threshold t, divide the image into two categories: target (gray level ≤ t) and background (gray level > t);

[0198] Calculate the pixel ratio of target and background:

[0199] ω0(t)=Σp(i), (i=0~t)

[0200] ω1(t)=Σp(i), (i=t+1~L-1)

[0201] Among them, p(i) is the probability of gray level i appearing, and L is the total number of gray levels;

[0202] Calculate the average gray level of the target and background:

[0203] μ0(t)=[Σi·p(i)] / ω0(t), (i=0~t)

[0204] μ1(t)=[Σi·p(i)] / ω1(t), (i=t+1~L-1)

[0205] Compute the overall average gray level of an image:

[0206] μ T =Σi·p(i), (i=0~L-1)

[0207] Calculate the between-class variance:

[0208] σ B 2 (t) = ω0(t) · [μ0(t) - μ T ] 2 +ω1(t)·[μ1(t)-μ T ] 2

[0209] Calculate the within-class variance:

[0210] σ W 2 (t) = ω0(t)·σ0 2 (t)+ω1(t)·σ1 2 (t)

[0211] Among them, σ0 2 (t) and σ1 2 (t) are the intra-class variances of target and background respectively;

[0212] 3) Select the best threshold: traverse all possible thresholds t and find the one that makes the inter-class variance σ B 2 (t) Maximum or intra-class variance σ W 2 (t) The smallest threshold is the optimal threshold;

[0213] Then, the grayscale Figure 2 Value, separate the water surface and fish body area, and complete the preprocessing of infrared thermal images;

[0214] Then, the preprocessed underwater image data is input into the pre-trained fish detection model to locate all sturgeons in each image and obtain the sturgeon location information;

[0215] In this embodiment, the fish body detection model is the YOLOv8 target detection algorithm, which can quickly and accurately locate the sturgeon in the image after a large number of sample training for sturgeons; during the training process, this embodiment uses a data set containing 10,000 labeled sturgeon images, of which 80% are used for training and 20% are used for verification; the training parameters are set as follows: batch size: 64; learning rate: initial value 0.01, using cosine annealing strategy; optimizer: AdamW, weight decay coefficient is 0.0005; training rounds: 100; the detection results include the position of the fish body (expressed in the form of a bounding box [x, y, w, h], where x, y are the center coordinates of the bounding box, w, h are the width and height), size (pixel area) and confidence (floating point number between 0 and 1);

[0216] Search for sturgeon fish bodies in the preprocessed infrared thermal image data according to the sturgeon position information, and accurately locate the fish body position in the thermal image in combination with the detection results of the underwater camera;

[0217] Specifically, the coordinate system mapping relationship between underwater image data and infrared thermal image data is first established using affine transformation, which is expressed as:

[0218]

[0219] Among them, (x i ,y i ) is the coordinate of the i-th pixel point in the underwater image data coordinate system, (x′ i ,y′ i ) is the corresponding pixel coordinate in the infrared thermal image data coordinate system, (x i ,y i ) and (x′ i ,y′ i ) constitute a set of pixel pairs; is the linear transformation matrix, is the translation matrix;

[0220] Use n sets of pixel pairs to construct the following linear equations:

[0221]

[0222] Wherein, n is a positive integer;

[0223] The constructed linear equations are solved using the least squares method to obtain the calculation results of the linear transformation matrix and the translation matrix. After obtaining the affine transformation matrix, the coordinates of any point in the underwater camera coordinate system can be converted into the coordinates of the corresponding point in the thermal imager coordinate system.

[0224] According to the calculation result, the sturgeon position information is mapped to the preprocessed infrared thermal imaging data, a plurality of mapping centers representing the sturgeon position are formed in the infrared thermal imaging data, and a search window is set with each mapping coordinate as the center, and a connected area with a temperature higher than a preset value is searched in the search window to obtain the corresponding sturgeon body;

[0225] Then, the body temperature characteristics of each sturgeon are extracted. For each located fish body area, the temperature values ​​of all pixels in the area are extracted; the average temperature in the area is calculated as the average body temperature of the fish, and the highest temperature value in the area is found as the highest body temperature of the fish; and the average temperature and standard deviation of all sturgeons in the school of fish are further calculated; the temperature change rate is calculated according to the average body temperature of the fish at each moment; the average body temperature and the temperature change rate are saved together as the body temperature characteristics of the sturgeon;

[0226] Determine whether the fish body temperature is abnormal based on the following principles:

[0227] 1) Absolute temperature threshold: set the upper and lower thresholds according to the normal body temperature range of sturgeons; in this embodiment, the normal body temperature range is set to 10°C to 25°C (may be adjusted according to the specific sturgeon species); low temperature threshold: T low =8℃; high temperature threshold: T high =28℃; when the average temperature is lower than the low temperature threshold or higher than the high temperature threshold, the fish body temperature is abnormal;

[0228] 2) Relative temperature difference: compare the temperature difference between different fish in the same pool; first calculate the mean μ and standard deviation σ of the average temperature of all fish in the pool, and set the abnormal judgment standard: |T-μ|>2σ. If this condition is met, the fish temperature is abnormal, where T is the average body temperature of a single fish;

[0229] 3) Temperature change trend: Analyze the temperature change trend of a single fish over time; record the temperature data of each fish in the past 24 hours, use linear regression to calculate the temperature change slope k, and set the slope threshold k threshold , when |k|>k threshold When , it is determined to be an abnormal change trend;

[0230] When any of the above conditions is met, the system marks the fish as having abnormal temperature and triggers the corresponding alarm;

[0231] Conducting behavior analysis on sturgeons based on detected sturgeon position information to obtain behavioral characteristics of sturgeons;

[0232] In this embodiment, the behavior analysis includes: swimming trajectory analysis, swimming speed analysis, posture analysis and group behavior analysis;

[0233] 1) Swimming trajectory analysis: Based on the position of the fish center at the previous moment, the Kalman filter algorithm is used to predict and update the position of the fish center at the next moment in real time, and obtain the movement trajectory of the fish center over a period of time; Kalman filtering is a recursive filter that can estimate dynamics from a series of measurements containing noise. Its core formula is as follows:

[0234] Prediction steps:

[0235]

[0236]

[0237] Update steps:

[0238]

[0239] P[k / k]=(1-K[k]·H k )·P[k / k-1]

[0240] in, is the state estimation, P is the error covariance matrix, F k is the state transfer matrix, Q k is the process noise covariance, H k is the observation matrix, R k is the observation noise covariance, K is the Kalman gain, and z is the observation value;

[0241] The steps to calculate the swimming trajectory are as follows:

[0242] a. Initialization:

[0243] State Vector Initialize the initial position and velocity of the center point of the fish body, for example [x position ,y position ,x velocity ,y velocity ];

[0244] Error covariance matrix P[0 / 0]: Initial uncertainty of the initialized state estimate, usually set to a diagonal matrix, where the diagonal elements represent the initial variance of the corresponding state variables;

[0245] State transition matrix F k : Describes the law of change of state variables over time, such as uniform linear motion model or uniform accelerated linear motion model;

[0246] Process noise covariance matrix Q k : represents the noise in the state transition process, such as water flow disturbance or the uncertainty of the fish's own movement;

[0247] The observation matrix Hk : describes the relationship between state variables and observations. For example, if the observation is the coordinate of the center point of the fish body, the observation matrix can be set to [1 0 0 0; 0 1 0 0];

[0248] Observation noise covariance matrix R k : represents the noise in the observation process, such as the error of the image recognition algorithm;

[0249] b. Prediction step (predict the current state based on the state estimate of the previous moment):

[0250] Status prediction Use the state transfer matrix and the state estimate of the previous moment to predict the state at the current moment;

[0251] Error Covariance Forecast Using the state transfer matrix, the error covariance of the previous moment and the process noise covariance, the error covariance of the current moment is predicted;

[0252] c. Update step (correct the state prediction based on the current observation value):

[0253] Kalman Gain Calculate the Kalman gain to balance the impact of predicted values ​​and observed values ​​on state estimation;

[0254] Status Update Use Kalman gain, observations, and state predictions to update the current state estimate;

[0255] Error covariance update P[k / k] = (1-K[k]·H k )·P[k / k-1]: Update the error covariance at the current moment using the Kalman gain and observation matrix;

[0256] d. Iteration:

[0257] Repeat steps b and c, and use the observed values ​​at each moment to continuously update the state estimation to obtain the motion trajectory of the center point of the fish body;

[0258] e. Output:

[0259] The final output is a series of state estimates Indicates the position and speed of the center point of the fish body at each moment. These points are connected to form the swimming trajectory of the fish body.

[0260] For example, if the coordinates of the center point of the fish body at three consecutive moments are observed to be (1,1), (2,2), and (3,3), the Kalman filter algorithm can be used to estimate the position and speed of the fish body at each moment and draw the swimming trajectory of the fish body; then the length, curvature and frequency of change of the movement direction of the movement trajectory can be further obtained;

[0261] Abnormal Behavior Assessment:

[0262] Track length: Healthy sturgeons usually swim continuously, with longer tracks; however, sick or injured sturgeons may swim weakly or stagnate, resulting in significantly shorter tracks.

[0263] Curvature of the track: The track of a healthy sturgeon is relatively smooth with less curvature; however, a sick sturgeon may have abnormal behaviors such as unstable swimming direction and spinning, which will increase the curvature of the track.

[0264] Frequency of trajectory direction changes: Healthy sturgeons swim in a relatively stable direction with a low frequency of direction changes; however, sick sturgeons may change their direction frequently, resulting in a higher frequency of direction changes;

[0265] 2) Swimming speed analysis:

[0266] The displacement of the fish body per unit time is calculated based on the coordinates of the center point of the fish body at several moments, and the instantaneous speed of the sturgeon is further calculated. The average speed over a period of time is calculated based on the instantaneous speed of the sturgeon, and the average speed is saved as the swimming speed feature to complete the swimming speed analysis;

[0267] Instantaneous speed calculation formula:

[0268] Average speed calculation formula:

[0269] in, is the coordinate of the center of the fish body in the i-th frame, Δt is the time interval between adjacent frames (times), and K is the total number of frames (times);

[0270] Abnormal behavior judgment: according to factors such as the species, size and breeding environment of the sturgeon, a reasonable swimming speed threshold is set; the average speed of each fish is compared with the preset threshold. If it is lower than the threshold, the fish is judged to be swimming slowly and may have health problems;

[0271] 3) Posture analysis:

[0272] The tilt angle of the fish body contour is used to determine whether there is a rollover or backstroke phenomenon, which is one of the obvious characteristics of sturgeon disease. The principal component analysis method is used to calculate the main axis direction of the fish body contour, specifically:

[0273] Calculate the covariance matrix Cov of the fish body contour points:

[0274]

[0275] Among them, (x j ,y j ) is the coordinate of the jth pixel in the fish body contour point; x and y are the mean values ​​of the horizontal and vertical coordinates of the fish body contour points respectively; the covariance matrix Cov of the fish body contour point is a 2×2 matrix used to describe the relationship between the coordinates x and y of the contour point; the upper left corner element Represents the variance of variable x; the lower right element Represents the variance of variable y; the lower left and upper right elements represents the covariance between variables x and y;

[0276] Construct the characteristic equation: Cov·v=λ·v, where λ and v are the eigenvalue and eigenvector of the covariance matrix Cov respectively;

[0277] Solving the characteristic equation to obtain a first eigenvalue λ1 and a first eigenvector v1 corresponding thereto, and a second eigenvalue λ2 and a second eigenvector v2 corresponding thereto;

[0278] The specific calculation process is as follows:

[0279] a. Calculate the characteristic polynomial:

[0280] det(Cov-λ·I)=0

[0281] Where I is the 2×2 identity matrix;

[0282] b. Solve the characteristic equation: Expand the characteristic polynomial to obtain a quadratic equation about λ. Solve the equation to obtain two eigenvalues ​​λ1 and λ2;

[0283] c. Calculate eigenvectors: Substitute each eigenvalue into the characteristic equation and solve the corresponding eigenvectors v1 and v2;

[0284] The eigenvector corresponding to the maximum eigenvalue is used as the main axis direction of the fish body contour, the main axis direction is used to calculate the inclination angle of the sturgeon swimming, and the inclination angle is saved as a posture feature to complete the posture analysis;

[0285] For example, suppose the covariance matrix Cov is: Then the characteristic polynomial is:

[0286] det(Cov-λ·I)=(4-λ)(3-λ)-2×2=λ 2 -7λ+8=0

[0287] Solving this equation yields two eigenvalues:

[0288]

[0289] Substituting λ1 into the characteristic equation, we obtain:

[0290]

[0291] Solving the system of equations, we obtain the eigenvector v1≈[0.85,0.53];

[0292] Similarly, substituting λ2 into the characteristic equation, we obtain the characteristic vector v2≈[-0.53,0.85];

[0293] The principal axis direction is the eigenvector corresponding to the maximum eigenvalue. In this example, λ1>λ2, so the principal axis direction is v1≈[0.85,0.53];

[0294] The calculation formula of the tilt angle θ is: θ = arctan (v y / v x )

[0295] Among them, (v x ,v y ) is the unit vector in the main axis direction; therefore, in this example, the sturgeon swimming tilt angle θ=arctan(0.53 / 0.85)≈31.9°;

[0296] Abnormal posture judgment:

[0297] Set the threshold angle θ threshold , when |θ|>θ threshold When the swimming posture is abnormal (such as sideways or backstroke),

[0298] 4) Group behavior analysis:

[0299] Analyze the relative positions and movement patterns of multiple fish and identify unusual flocking or dispersal behaviors using the following methods:

[0300] Movement consistency analysis: Calculate the angle between the average velocity vector of the fish school and the velocity vector of each individual, and use the instantaneous speed of each sturgeon to calculate the consistency index C of the fish school:

[0301]

[0302] Wherein, N is the number of sturgeons in the fish school; the larger the consistency index C is, the more consistent the movement of the fish school is, and the closer the C value is to 0, the more dispersed the movement of the group is; in this embodiment, a threshold h of the consistency index is set, and h=0.3;

[0303] Spatial distribution analysis: The nearest neighbor distance (NND) method was used to evaluate the spatial distribution of fish schools. The Euclidean distance between each sturgeon and its nearest neighbor was calculated to obtain the nearest neighbor distance of each sturgeon, and the average nearest neighbor distance of the fish school was calculated.

[0304] The details are as follows:

[0305] a. Data preparation:

[0306] First, you need to obtain the coordinate information of each fish in space, such as (x, y) coordinates;

[0307] Store the coordinate information of each fish in a data structure, such as a list or array, to facilitate subsequent calculations;

[0308] b. Traverse each fish: Use a loop to traverse each fish and take it as the target fish;

[0309] c. Calculate the distance between the target fish and other fish:

[0310] Inside the loop, iterate over all the fish again (including the target fish itself);

[0311] To calculate the distance between the target fish and the currently traversed fish, you can use the Euclidean distance formula:

[0312] Euclidean distance = sqrt[(x1-x2) 2 +(y1-y2) 2 ]

[0313] Among them, (x1, y1) is the coordinate of the target fish, and (x2, y2) is the coordinate of the currently traversed fish;

[0314] d. Find the nearest neighbor fish:

[0315] In the process of calculating the distance, the minimum distance between the target fish and other fish and the corresponding fish number or index are recorded;

[0316] Ignore the target fish itself, that is, the minimum distance cannot be 0;

[0317] e. Store the nearest neighbor distance:

[0318] Store the found minimum distance in a list or array, which is the NND distribution;

[0319] f. Continue to traverse the next fish and repeat steps be until all fish have been traversed;

[0320] Final result: a list / array containing the nearest neighbor distances of all fish, i.e. NND distribution; and the average nearest neighbor distance of a fish group = the sum of the nearest neighbor distances of all fish / the number of fish;

[0321] For example, suppose there are 4 fish with coordinates (1,2), (3,4), (5,1), and (2,3).

[0322] The calculation process is as follows:

[0323] Fish 1(1,2):

[0324] Calculate the distance between fish 1 and fish 1: sqrt((1-1) 2 +(2-2) 2 )=0;

[0325] Calculate the distance between fish 1 and fish 2: sqrt((1-3) 2 +(2-4) 2 ) = sqrt(8) ≈ 2.83;

[0326] Calculate the distance between fish 1 and fish 3: sqrt((1-5) 2 +(2-1) 2 ) = sqrt(17) ≈ 4.12;

[0327] Calculate the distance between fish 1 and fish 4: sqrt((1-2) 2 +(2-3) 2 ) = sqrt(2) ≈ 1.41;

[0328] The nearest neighbor fish is fish 4, and the nearest neighbor distance is 1.41;

[0329] Fish 2 (3,4):

[0330] Calculate the distance between fish 2 and fish 1: sqrt((3-1) 2 +(4-2) 2 ) = sqrt(8) ≈ 2.83;

[0331] Calculate the distance between fish 2 and fish 2: sqrt((3-3) 2 +(4-4) 2 )=0;

[0332] Calculate the distance between fish 2 and fish 3: sqrt((3-5) 2 +(4-1) 2 ) = sqrt(13) ≈ 3.61;

[0333] Calculate the distance between fish 2 and fish 4: sqrt((3-2) 2 +(4-3) 2) = sqrt(2) ≈ 1.41;

[0334] The nearest neighbor fish is fish 4, and the nearest neighbor distance is 1.41;

[0335] Fish 3 (5,1):

[0336] Calculate the distance between fish 3 and fish 1: sqrt((5-1) 2 +(1-2) 2 ) = sqrt(17) ≈ 4.12;

[0337] Calculate the distance between fish 3 and fish 2: sqrt((5-3) 2 +(1-4) 2 ) = sqrt(13) ≈ 3.61;

[0338] Calculate the distance between fish 3 and fish 3: sqrt((5-5) 2 +(1-1) 2 )=0;

[0339] Calculate the distance between fish 3 and fish 4: sqrt((5-2) 2 +(1-3) 2 ) = sqrt(13) ≈ 3.61;

[0340] The nearest neighbor fish is fish 2 or fish 4, and the nearest neighbor distance is 3.61;

[0341] Fish 4 (2,3):

[0342] Calculate the distance between fish 4 and fish 1: sqrt((2-1) 2 +(3-2) 2 ) = sqrt(2) ≈ 1.41;

[0343] Calculate the distance between fish 4 and fish 2: sqrt((2-3) 2 +(3-4) 2 ) = sqrt(2) ≈ 1.41;

[0344] Calculate the distance between fish 4 and fish 3: sqrt((2-5) 2 +(3-1) 2 ) = sqrt(13) ≈ 3.61;

[0345] Calculate the distance between fish 4 and fish 4: sqrt((2-2) 2 +(3-3) 2 )=0;

[0346] The nearest neighbor fish is fish 1 or fish 2, and the nearest neighbor distance is 1.41;

[0347] The final NND distribution is [1.41, 1.41, 3.61, 1.41];

[0348] In this example, the sum of the nearest neighbor distances of all fish is the sum of all distances in the NND distribution: 1.41+1.41+3.61+1.41=7.84, and the number of fish is 4; the sum of the nearest neighbor distances of all fish is divided by the number of fish to get the average nearest neighbor distance of the fish group: 7.84 / 4=1.96;

[0349] In the above example, the nearest neighbor distance of fish 3 (5, 1) is 3.61, which is greater than the average nearest neighbor distance of the fish group, 1.96, and is an abnormal object that needs attention;

[0350] Group behavior analysis can be used to identify abnormal dispersion of fish schools in a pond and isolated behavior of sick fish; healthy sturgeons usually move in groups, while sick or injured sturgeons may leave the school, move alone or avoid other fish; based on the above analysis, this embodiment sets the following abnormal behavior judgment criteria:

[0351] If the swimming direction and speed of a fish are significantly different from the average direction and speed of the school of fish, it may be isolated and out of the group; when the consistency index C < 0.3, it is judged as abnormal dispersion; if the nearest neighbor distance of a fish is significantly greater than the average nearest neighbor distance of the school of fish (for example, the former is more than three times the latter), it may be isolated and out of the group, and it is judged as abnormal spatial distribution;

[0352] Then, multi-source data fusion is performed, and the sturgeon behavior characteristics and fish body temperature characteristics are fused according to the following formula to obtain the sturgeon fusion feature F fused :

[0353]

[0354] Among them, f1~f s Respectively represent the 1st to sth characteristic index values ​​of sturgeon; f 1,max ~f s,max They represent the maximum values ​​of the 1st to sth characteristic index values ​​of sturgeon respectively; f 1,min ~f s,min represent the minimum values ​​of the 1st to sth characteristic index values ​​of sturgeon; w1~w s They represent the weight factors corresponding to the 1st to sth characteristic index values ​​of sturgeon respectively; this fusion method comprehensively considers the two factors of weight distribution and data normalization, and fuses multiple variables together through linear combination; the specific values ​​of each feature are mapped to the interval [0,1] respectively, realizing data normalization; this makes data in different ranges comparable, avoiding the fusion result dominated by data in a larger range;

[0355] The characteristic indicators of the sturgeon include: any one of the length of the motion trajectory, curvature, frequency of motion direction change, average speed, tilt angle, nearest neighbor distance, average temperature of the fish body and temperature change rate;

[0356] Finally, the sturgeon fusion features are input into the pre-trained anomaly detection model to obtain the anomaly detection results and complete the intelligent monitoring of sturgeon health;

[0357] In this embodiment, the anomaly detection model is specifically a multi-layer LSTM (Long Short-Term Memory) neural network model; the multi-layer LSTM can learn more complex time series features, thereby improving the accuracy and robustness of the model; the multi-layer LSTM stacks multiple LSTM layers together, and the output of each layer serves as the input of the next layer; this structure can learn time series features at different levels, where the bottom layer learns short-term features and the high layer learns long-term features;

[0358] The multi-layer LSTM neural network model includes: an input layer, a plurality of LSTM layers connected in sequence, a fully connected layer and an output layer; the input layer is used to receive the sturgeon fusion feature F fused ,The LSTM layer structure is the same, both are set with 128 units, using the tanh activation function and a Dropout rate of 0.2; the fully connected layer includes 32 neurons and uses the ReLU activation function; the output layer includes 1 neuron and uses the Sigmoid activation function to output anomaly scores between 0 and 1;

[0359] A sliding window is set with a window size of 60 (corresponding to 1 minute of data). The sturgeon fusion features obtained in the historical time period are used to generate time series samples, which are divided into training set, validation set and test set at a ratio of 8:1:1. The multi-layer LSTM neural network model is supervised by the binary cross entropy function for training. The optimizer is Adam, the initial value of the learning rate is 0.001, and the learning rate decay strategy is used to prevent the model from overfitting. The training batch size is 64, the training rounds are 100, and the early stopping strategy is used to avoid overfitting. The training is stopped when the validation set loss does not decrease for 10 consecutive rounds. The model is tested using the test set. When the test indicators meet expectations, the trained anomaly detection model is obtained.

[0360] Input the sturgeon fusion features obtained in real time into the trained anomaly detection model to obtain anomaly detection results, and issue an anomaly alarm based on the anomaly detection results; the anomaly detection results are the anomaly scores of each feature index of the sturgeon in each time window;

[0361] For the data of each time window, the model outputs an anomaly score z between 0 and 1, indicating the probability of an anomaly occurring within that time window; the anomaly score threshold is set to 0.7, and when the anomaly score 0.7 < z, it is determined as an anomaly;

[0362] When an abnormal state of the sturgeon is detected, generate an alarm at the following levels and send it to the smart terminal of the management personnel:

[0363] 1) Minor alarm:

[0364] Trigger condition: Slight anomaly in a single indicator, such as the behavior of a single fish (slow swimming speed or short swimming trajectory or the nearest neighbor distance of a single fish is significantly greater than the average nearest neighbor distance of the fish school) or the temperature is slightly deviated from the normal range or the output anomaly scores z of a small number of intelligent analysis modules (comprehensive indicators) are greater than the set threshold;

[0365] Alarm method: Display a yellow warning icon on the system interface and send a push notification to the management personnel;

[0366] Suggested operation: Closely monitor the subsequent state of the fish;

[0367] 2) Moderate alarm:

[0368] Trigger condition: Anomaly in multiple indicators or severe anomaly in a single indicator, such as abnormal behaviors (slow swimming) of multiple fish at the same time or abnormal group spatial distribution (consistency index C < 0.3), or a single fish is overturned or swimming on its back, or the body temperature of a single fish is significantly high or multiple anomaly score z values are greater than the set threshold;

[0369] Alarm method: Display an orange warning icon on the system interface and send a mobile push notification to the management personnel;

[0370] Suggested operation: Check the relevant fish and water quality parameters in a timely manner, and conduct isolation observation and treatment if necessary;

[0371] 3) Severe alarm:

[0372] Trigger condition: Continuous anomaly or severe anomaly in multiple indicators, such as abnormal behaviors (slow swimming and overturned or swimming on its back) of a large area of fish school or abnormal temperature, or a large number of anomaly scores z are greater than the set threshold or symptoms of suspected infectious diseases are detected;

[0373] Alarm method: Display a red warning icon on the system interface and send text messages and phone notifications to all relevant personnel;

[0374] Suggested operation: Immediately take emergency measures, such as isolating the affected fish school, adjusting water quality parameters or starting the disease prevention and control process, etc.;

[0375] In addition, in this embodiment, the model can be retrained regularly (e.g., weekly) using newly collected data to adapt to changes in the environment and the state of the fish school, while implementing an incremental learning mechanism to allow the model to adapt to new patterns without losing the learned knowledge;

[0376] It is worth mentioning that this embodiment can use the historical data accumulated over a long period of time for mining and analysis to discover potential laws, such as: disease prediction: analyzing the laws of disease occurrence in historical data and predicting the risk of diseases that may occur in the future; breeding strategy optimization: analyzing the impact of different breeding strategies on fish growth and health and optimizing breeding strategies; water quality parameter analysis: analyzing the relationship between water quality parameters and fish health and formulating more scientific water quality management plans;

[0377] The sturgeon health intelligent monitoring method of this embodiment was field tested for 6 months in a large sturgeon farm. The farm has 20 circular breeding ponds, each with a diameter of 15 meters and a water depth of 2 meters, with a total breeding capacity of about 100,000 sturgeons. The test results show:

[0378] 1) Anomaly detection accuracy: The accuracy of behavioral anomaly detection reached 94.3%, the accuracy of body temperature anomaly detection reached 96.7%, and the accuracy of comprehensive anomaly detection reached 95.8%;

[0379] 2) Response time: The average time from when an abnormality occurs to when the system issues an alarm is reduced to less than 2 minutes, and the average time from when the alarm is issued to when the farm staff arrives at the scene is reduced to less than 10 minutes;

[0380] 3) Disease prevention effect: The early disease identification rate increased by 78%, greatly reducing the risk of large-scale disease outbreaks, and the overall mortality rate of fish stocks decreased by 35%;

[0381] 4) Improved production efficiency: Feed utilization rate increased by 12%, mainly due to more accurate feeding time and quantity control; fish growth rate increased by 8% on average, shortening the breeding cycle;

[0382] 5) Economic benefits: The use of this method has increased the annual profit of the farm by about 20%, and the payback period is expected to be 1.5 years;

[0383] 6) Management efficiency: The daily inspection time of aquaculture management personnel has been reduced by 60%, allowing them to focus more on optimizing aquaculture strategies. The processing time of abnormal events has been shortened by an average of 45%;

[0384] 7) Data value: The large amount of data accumulated by the system provides a scientific basis for the farm to formulate long-term development strategies. Based on historical data analysis, the farm successfully predicted and responded to several potential water quality crises;

[0385] 8) Environmental impact: Due to more precise feeding and water quality management, the water quality of the breeding ponds has improved significantly, with ammonia nitrogen and nitrite content reduced by 25% and 30% respectively; the water consumption of the farm has been reduced by 15%, reducing the pressure on surrounding water resources;

[0386] 9) Scalability: This method was successfully adapted to different sizes of breeding ponds and different species of sturgeons, showing good versatility. During the test period, sensors such as dissolved oxygen and pH were also successfully integrated to further improve the comprehensiveness of monitoring;

[0387] This method realizes comprehensive, real-time and intelligent monitoring of the sturgeon farming process by integrating multiple sensor technologies and advanced data analysis algorithms. It can not only detect and handle abnormal situations in a timely manner, but also provide scientific basis and optimization suggestions for farming management through long-term data accumulation and analysis.

[0388] The implementation of this method significantly improves the efficiency and product quality of sturgeon farming, reduces disease risks and economic losses, and also provides new ideas and methods for the intelligent and refined management of the aquaculture industry. With the continuous advancement of technology and continuous optimization of the system, the present invention is expected to play an important role in a wider range of aquaculture fields and promote the entire industry to develop in a more efficient and sustainable direction.

[0389] The same or similar reference numerals correspond to the same or similar components;

[0390] The terms used to describe the positional relationship in the drawings are only used for illustrative purposes and should not be construed as limiting the present invention;

[0391] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. A sturgeon health intelligent monitoring method based on multi-source data fusion, characterized in that: The following steps are involved: S1: Real-time synchronous acquisition of underwater image data and infrared thermal image data of sturgeon breeding ponds, and pre-processing them respectively; S2: inputting the pre-processed underwater image data into a pre-trained fish body detection model, locating all sturgeons in each image, obtaining the sturgeon position information, and searching for sturgeon fish bodies in the pre-processed infrared thermal image data according to the sturgeon position information; S3: Analyze the behavior of the sturgeon according to the position information of the sturgeon to obtain the behavior characteristics of the sturgeon; The behavior analysis includes: swimming trajectory analysis, swimming speed analysis, posture analysis and group behavior analysis; Extracting the body temperature characteristics of each sturgeon from the preprocessed infrared thermal imaging data; S4: fusing the sturgeon behavior characteristics and the fish body temperature characteristics to obtain sturgeon fusion characteristics; S5: Input the sturgeon fusion features into a pre-trained anomaly detection model to obtain anomaly detection results, and complete intelligent monitoring of sturgeon health.

2. The method for intelligent monitoring of sturgeon health based on multi-source data fusion according to claim 1 is characterized in that: In step S1, the following operations are performed on the underwater image data to obtain pre-processed underwater image data: Image denoising: Use Gaussian filtering or median filtering to remove noise caused by the underwater environment; Contrast enhancement: The contrast-limited adaptive histogram equalization algorithm is used to improve the clarity of underwater image data; Color correction: Select a reference white point from the underwater image data, calculate the average value of the reference white point on the three color channels of red, green and blue, and compare it with the preset standard values ​​of the three color channels to obtain the gain coefficient of each color channel; use the gain coefficients of the three color channels of red, green and blue to adjust the white balance of the underwater image data; Perform the following operations on the infrared thermal imaging data to obtain pre-processed infrared thermal imaging data: Image segmentation: convert the infrared thermal image data into a grayscale image, use the OTUS algorithm to calculate the segmentation threshold of the water surface and the fish body area, binarize the grayscale image according to the segmentation threshold, and separate the water surface and the fish body area.

3. The method for intelligent monitoring of sturgeon health based on multi-source data fusion according to claim 1 is characterized in that: In the step S2, the fish body detection model is specifically: a YOLOv8 target detection model, and the sturgeon position information includes: the coordinates of the center point of the sturgeon body and the contour points of the sturgeon body.

4. The method for intelligent monitoring of sturgeon health based on multi-source data fusion according to claim 1 is characterized in that: In the step S2, searching for the sturgeon body in the pre-processed infrared thermal imaging data according to the sturgeon position information includes the following steps: Affine transformation is used to establish the coordinate system mapping relationship between underwater image data and infrared thermal imaging data, which can be expressed as: Among them, (x i ,y i ) is the coordinate of the i-th pixel point in the underwater image data coordinate system, (x i ′,y i ′) is the corresponding pixel coordinate in the infrared thermal image data coordinate system, (x i ,y i ) and (x i ′,y i ′) constitute a set of pixel pairs; is the linear transformation matrix, is the translation matrix; Use n sets of pixel pairs to construct the following linear equations: Wherein, n is a positive integer; The constructed linear equations are solved using the least squares method to obtain the calculation results of the linear transformation matrix and the translation matrix; According to the calculation result, the sturgeon position information is mapped to the preprocessed infrared thermal imaging data, and a plurality of mapping centers representing the sturgeon positions are formed in the infrared thermal imaging data. A search window is set with each mapping coordinate as the center, and a connected area with a temperature higher than a preset value is searched in the search window to obtain the corresponding sturgeon body.

5. The method for intelligent monitoring of sturgeon health based on multi-source data fusion according to claim 4 is characterized in that: In step S3, extracting the body temperature characteristics of each sturgeon includes: In the pre-processed infrared thermal imaging data, for each located sturgeon body area, the temperature values ​​of all pixels in the area are extracted, the temperature mean of all pixels in the area is calculated as the average body temperature of the corresponding sturgeon, and the average temperature and standard deviation of all sturgeons in the school of fish are further calculated; Calculate the temperature change rate based on the average temperature of the fish body at each moment; The fish body average temperature and the temperature change rate are stored together as the fish body temperature characteristics of the sturgeon.

6. The method for intelligent monitoring of sturgeon health based on multi-source data fusion according to claim 3 is characterized in that: In step S3, the swimming trajectory analysis includes: Based on the position of the fish center at the previous moment, the Kalman filter algorithm is used to predict and update the position of the fish center at the next moment in real time, obtain the movement trajectory of the fish center over a period of time, and further obtain the length, curvature and movement direction change frequency of the movement trajectory. The length, curvature and movement direction change frequency of the movement trajectory are saved together as the swimming trajectory features to complete the swimming trajectory analysis; The swimming speed analysis includes: The displacement of the fish body per unit time is calculated based on the coordinates of the center point of the fish body at several moments, and the instantaneous speed of the sturgeon is further calculated. The average speed over a period of time is calculated based on the instantaneous speed of the sturgeon, and the average speed is saved as the swimming speed feature to complete the swimming speed analysis; The posture analysis includes: The principal component analysis method is used to calculate the principal axis direction of the fish body contour, specifically: Calculate the covariance matrix Cov of the fish body contour points: Among them, (x j ,y j ) is the coordinate of the jth pixel point in the fish body contour point; and are the mean values ​​of the horizontal and vertical coordinates of the fish body contour points respectively; Construct the characteristic equation: Cov·v=λ·v, where λ and v are the eigenvalue and eigenvector of the covariance matrix Cov respectively; Solving the characteristic equation to obtain a first eigenvalue λ1 and a first eigenvector v1 corresponding thereto, and a second eigenvalue λ2 and a second eigenvector v2 corresponding thereto; The eigenvector corresponding to the maximum eigenvalue is used as the main axis direction of the fish body contour, the main axis direction is used to calculate the inclination angle of the sturgeon swimming, and the inclination angle is saved as a posture feature to complete the posture analysis; The group behavior analysis includes: Movement consistency analysis: The instantaneous speed of each sturgeon is used to calculate the consistency index C of the fish school: Among them, u k is the instantaneous speed of the sturgeon at the kth moment; N is the number of sturgeons in the school of fish; the larger the consistency index C is, the more consistent the movement of the school of fish is; Spatial distribution analysis: The spatial distribution of fish schools was evaluated using the nearest neighbor distance method, the Euclidean distance between each sturgeon and its nearest neighbor was calculated, the nearest neighbor distance of each sturgeon was obtained, and the average nearest neighbor distance of the fish school was calculated; The nearest neighbor distance of each sturgeon is saved as a group behavior feature to complete the group behavior analysis; The swimming trajectory characteristics, swimming speed characteristics, posture characteristics and group behavior characteristics are stored together as the sturgeon behavior characteristics.

7. A sturgeon health intelligent monitoring method based on multi-source data fusion according to claims 5 and 6, characterized in that: In step S4, the sturgeon behavior characteristics and the fish body temperature characteristics are fused according to the following formula to obtain the sturgeon fusion feature F fused : Among them, f1~f s Respectively represent the 1st to sth characteristic index values ​​of sturgeon; f 1,max ~f s,max They represent the maximum values ​​of the 1st to sth characteristic index values ​​of sturgeon respectively; f 1,min ~f s,min represent the minimum values ​​of the 1st to sth characteristic index values ​​of sturgeon; w1~w s They represent the weight factors corresponding to the 1st to sth characteristic index values ​​of sturgeon respectively; The characteristic indicators of the sturgeon include: any one of the length of the motion trajectory, the curvature, the frequency of change of the motion direction, the average speed, the tilt angle, the nearest neighbor distance, the average temperature of the fish body and the temperature change rate.

8. The method for intelligent monitoring of sturgeon health based on multi-source data fusion according to claim 7 is characterized in that: In step S5, the anomaly detection model is specifically a multi-layer LSTM neural network model; The multi-layer LSTM neural network model includes: an input layer, a plurality of sequentially connected LSTM layers, a fully connected layer, and an output layer; A sliding window is set, and a time series sample is generated using the sturgeon fusion features obtained in the historical time period, and the samples are divided into a training set, a validation set, and a test set. The multi-layer LSTM neural network model is supervised for training using a binary cross entropy function, and a trained anomaly detection model is obtained after verification and testing; The sturgeon fusion features obtained in real time are input into the trained anomaly detection model to obtain anomaly detection results, and anomaly alarm is performed according to the anomaly detection results; the anomaly detection results are the anomaly scores of each feature index of the sturgeon in each time window.

9. The method for intelligent monitoring of sturgeon health based on multi-source data fusion according to claim 8, characterized in that: The trained anomaly detection model determines whether each feature indicator is abnormal based on the following rules: If the length of the sturgeon's movement trajectory is less than the first preset threshold, the sturgeon's behavior is abnormal; If the curvature of the sturgeon's motion trajectory is greater than a second preset threshold, the sturgeon's behavior is abnormal; If the frequency of the sturgeon's movement direction changes is greater than a third preset threshold, the sturgeon's behavior is abnormal; If the average speed of the sturgeon is less than the fourth preset threshold, the sturgeon behaves abnormally; If the sturgeon's tilt angle is greater than a fifth preset threshold, the sturgeon's behavior is abnormal; If the absolute value of the difference between the nearest neighbor distance of the sturgeon and the average nearest neighbor distance of the fish school is greater than a sixth preset threshold, the sturgeon behaves abnormally; If any of the following conditions occur, the sturgeon temperature is abnormal: The average body temperature of the sturgeon exceeds the preset normal temperature range, the absolute value of the difference between the average body temperature of the sturgeon and the average temperature of all sturgeons in the school is greater than two standard deviations, and the temperature change rate is greater than the seventh preset threshold.

10. A sturgeon health intelligent monitoring system based on multi-source data fusion, using a sturgeon health intelligent monitoring method based on multi-source data fusion as described in any one of claims 1 to 9, characterized in that: include: Preprocessing unit: used to synchronously collect underwater image data and infrared thermal image data of sturgeon breeding ponds in real time and perform preprocessing respectively; A fish body positioning unit is used to input the pre-processed underwater image data into a pre-trained fish body detection model, locate all sturgeons in each image, obtain the sturgeon position information, and search for sturgeons in the pre-processed infrared thermal image data according to the sturgeon position information; Feature extraction unit: used for performing behavior analysis on the sturgeon according to the sturgeon position information to obtain the behavior characteristics of the sturgeon; The behavior analysis includes: swimming trajectory analysis, swimming speed analysis, posture analysis and group behavior analysis; and for extracting the body temperature characteristics of each sturgeon in the pre-processed infrared thermal imaging data; Feature fusion unit: used for fusing the sturgeon behavior feature and the fish body temperature feature to obtain the sturgeon fusion feature; Anomaly detection unit: used to input the sturgeon fusion features into a pre-trained anomaly detection model, obtain anomaly detection results, and complete intelligent monitoring of sturgeon health.

Images