A method for assessing shrimp health status based on multi-source information fusion

Abstract

This invention discloses a method for assessing shrimp health status based on multi-source information fusion, which relates to the field of data processing technology. The method includes the following steps: acquiring and preprocessing shrimp images at different growth stages; generating training images using artificial labels; extracting features such as shrimp outline length, area, and shell contrast; establishing a target recognition and feature extraction model based on the training images; identifying shrimp in the monitored aquaculture environment and extracting their behavioral characteristic parameters after training; calculating the behavioral assessment influence coefficient and growth assessment influence coefficient; and obtaining water source characteristic parameters to calculate the disease incidence influence coefficient; finally, generating a comprehensive health status assessment index, comparing it with a health judgment threshold, and outputting the health status judgment result. This achieves effective assessment of shrimp health status, real-time monitoring, and early warning, helping farmers take timely measures to prevent disease occurrence and improve shrimp survival rate and economic benefits.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to a method for assessing the health status of shrimp based on multi-source information fusion. Background Technology

[0002] In the global aquaculture industry, shrimp, as a major economic aquatic product, has received widespread attention. With advancements in aquaculture technology, the scale of shrimp farming has continuously expanded; however, at the same time, the health management issues encountered during the farming process have become increasingly prominent. In particular, assessing the growth status and health condition of shrimp, and the impact of environmental factors, has become a crucial issue for increasing yields and ensuring aquatic product safety. Traditional shrimp health assessment methods largely rely on human experience and simple growth indicators, lacking systematicity and scientific rigor, and are easily influenced by subjective factors, leading to inaccurate and unstable assessment results.

[0003] In recent years, the rapid development of computer vision and deep learning technologies has brought new opportunities for health management in aquaculture. By effectively fusing multi-source information, more accurate assessments of shrimp health can be achieved. Image processing technology can provide detailed information on shrimp's physical characteristics, while environmental monitoring data provides contextual information for health assessment. However, how to efficiently integrate this diverse information and analyze and judge it using scientific methods remains a major challenge in the current technological field.

[0004] Especially in shrimp health monitoring, it is often necessary to consider both biological characteristics and environmental factors simultaneously. For example, shrimp swimming behavior, body shape, and water quality indicators in the external environment can all have a profound impact on their health status. Traditional methods often fail to comprehensively consider these factors, leading to a one-sided assessment of shrimp health status. Furthermore, how to acquire and process data in real time and accurately in response to the dynamic changes of various environmental and biological indicators is a problem that current technology urgently needs to solve.

[0005] In the prior art, CN119474949A discloses a method for assessing shrimp population health status based on deep clustering analysis. This method includes: monitoring health parameters; determining a dispersed grid of shrimp populations; determining a temporary grid; identifying abnormal grids outside the shrimp population; calculating the actual health index; adjusting the grid step size; and assessing health. This invention, by combining real-time physiological and behavioral data such as density, saturation, and activity levels, can accurately monitor and assess the health status of shrimp populations. By progressively eliminating abnormal grids, more representative grids are selected, avoiding assessment biases caused by environmental fluctuations or measurement errors. This not only achieves dynamic monitoring of shrimp population health status but also automatically adjusts the grid step size by comparing the ideal health index with the actual health index, thereby improving the accuracy and reliability of health assessment. However, shrimp population health status is affected by various biological factors, including physiological state, disease, and nutritional levels. Although this method focuses on physiological and behavioral data for assessment, other uncontrollable biological factors may be overlooked in the assessment, affecting the comprehensive evaluation of health status. Therefore, the accuracy and effectiveness of the assessment results are reduced.

[0006] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0007] The purpose of this invention is to provide a method for assessing the health status of shrimp based on multi-source information fusion, so as to solve the problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention provides the following technical solution:

[0009] A method for assessing shrimp health status based on multi-source information fusion, comprising the following steps:

[0010] Several images of shrimp at different growth stages were acquired. The acquired shrimp images were preprocessed. Based on the preprocessed shrimp images, each shrimp target in the preprocessed shrimp images was labeled by a label box using a manual labeling method. The image feature parameters corresponding to the labeled shrimp in the image were obtained, and the labeled images were recorded as training images. The image feature parameters include shrimp outline length, shrimp outline area, and shrimp shell contrast.

[0011] Based on the obtained training images, a target recognition model and a feature extraction model are established. The training images are used as input to the target recognition model, and the corresponding bounding boxes in the training images are used as labels to train the target recognition model. The shrimp region images identified by the target recognition model are cropped and used as input to the feature extraction model, and the corresponding image feature parameters are used as labels to train the feature extraction model.

[0012] Continuous time-series images of the shrimp farming environment to be monitored were collected during the collection period. After preprocessing, the continuous time-series images were input into a trained target recognition model to identify shrimp targets in the images and add bounding boxes to them. The images of the regions within the bounding boxes were then input into a trained feature extraction model to obtain the image feature parameters corresponding to the shrimp targets. Based on the generated bounding boxes and the image feature parameters corresponding to the shrimp targets, the behavioral feature parameters of each shrimp target during the collection period were extracted using the DeepSORT algorithm.

[0013] Based on the obtained behavioral and image feature parameters of the shrimp target, behavioral assessment influence coefficients and growth assessment influence coefficients are calculated and generated. At the same time, water source feature parameters of the shrimp farming environment to be monitored during the collection period are obtained, and disease incidence influence coefficients are calculated based on the water source feature parameters. The behavioral feature parameters include shrimp swimming displacement and swimming frequency, and the water source feature parameters include total bacterial count, number of pathogenic Vibrio bacteria, water temperature, and water pH value.

[0014] Based on the obtained disease incidence impact coefficient, behavior assessment impact coefficient, and growth assessment impact coefficient, a health status assessment index is generated. The generated health status assessment index is compared with the set health judgment threshold. Based on different comparison results, the corresponding health status judgment result is issued to complete the shrimp health status assessment.

[0015] Furthermore, several shrimp images at different growth stages are acquired, and the acquired shrimp images are preprocessed. The preprocessing includes image enhancement and denoising preprocessing. Wavelet transform is used to denoise each shrimp image, and bilateral filtering is used to enhance each shrimp image.

[0016] Each shrimp target in the preprocessed shrimp image is labeled using a manual labeling method. The specific labeling method is as follows: each shrimp target in the image is labeled using an image labeling tool, and the labeling box is the minimum bounding rectangle of each shrimp target.

[0017] Furthermore, based on the obtained training images, a target recognition model is established, which is based on the YOLOv5s network model. The trained target recognition model is used to identify and mark all shrimp targets in the shrimp image and extract the marked area image.

[0018] A feature extraction model is established based on a convolutional neural network. This convolutional neural network consists of an input layer, convolutional layers, pooling layers, fully connected layers, and an output layer. The activation function in the convolutional layers is the ReLU function, and the specific expression of the ReLU function is as follows:

[0019] ReLU(s l(p,q) ) = max(0,s l(p,q) )

[0020] Where l represents the l-th corresponding convolutional layer, s l(p,q) Then it represents the q-th feature value of the p-th training sample image in the l-th corresponding convolutional layer, where l is the index of the convolutional layer, p is the index of the training sample image, and q is the index of the feature value in the training sample image;

[0021] For the fully connected layer, the number of neurons in the fully connected layer is set to 32, the initial neural network learning rate is set to 0.001, and the number of training rounds is 200.

[0022] The trained feature extraction model takes the labeled region image as input and outputs the corresponding predicted image feature parameters.

[0023] Furthermore, based on the obtained behavioral and image feature parameters of the shrimp target, behavioral assessment influence coefficients and growth assessment influence coefficients are calculated and generated. The formula for calculating the behavioral assessment influence coefficient is as follows:

[0024]

[0025] In the formula, CIF is the behavioral assessment impact coefficient, and V mean f represents the average swimming speed of all identified shrimp targets within the monitored shrimp farming environment during the data collection period. mean V0 represents the average number of swims of all identified shrimp targets within the monitored shrimp farming environment during the data collection period. V0 represents the set reference value for swimming speed, and f0 represents the set reference value for the number of swims. The average speed V0 of each swim during the data collection period is also represented by the value. mean The calculation is performed based on the displacement and time of each swim, using the following formula:

[0026]

[0027] In the formula, V u Let be the average swimming speed of the u-th shrimp during each swimming period, where u is the index of the identified shrimp target, u∈[1,U], and U is the total number of identified shrimp targets.

[0028]

[0029] In the formula, X u,f Let Δt be the displacement of the u-th shrimp target during its f-th swim. u,fLet f be the time interval of the u-th shrimp target during the f-th swim, where f is the index of the number of swims of the u-th shrimp target during the collection time period, F is the total number of swims of the u-th shrimp target during the collection time period, and f∈[1,F].

[0030] The specific formula used to calculate the growth assessment impact coefficient is as follows:

[0031]

[0032] In the formula, ECI is the growth assessment influence coefficient, and AC mean L represents the average shell contrast of all identified shrimp targets within the monitored shrimp farming environment. mean M represents the average length of the outlines of all shrimp targets identified within the shrimp farming environment to be monitored. mean L0 represents the average area of ​​all shrimp target outlines identified within the shrimp farming environment to be monitored, while M0 represents the reference values ​​for the shrimp target outline length and shrimp target outline area.

[0033] Furthermore, water source characteristic parameters of the shrimp farming environment to be monitored were simultaneously acquired during the collection period. Based on these water source characteristic parameters, the disease incidence impact coefficient was calculated. The specific formula used to calculate the disease incidence impact coefficient is as follows:

[0034]

[0035] In the formula, SCI is the disease incidence impact coefficient, RE is the total number of bacterial colonies in the water, HZ is the number of pathogenic Vibrio, and pH is... S The pH value of the water body, T S T0 represents the water temperature, PH0 represents the reference value for water pH, and T0 represents the reference value for water temperature.

[0036] Furthermore, based on the obtained disease incidence impact coefficient, behavioral assessment impact coefficient, and growth assessment impact coefficient, a health status assessment index is generated. The specific formula used to calculate the health status assessment index is as follows:

[0037]

[0038] In the formula, ZH is the health status assessment index, and ω1, ω2 and ω3 are the weight coefficients of the disease incidence influence coefficient, the behavioral assessment influence coefficient and the growth assessment influence coefficient, respectively, where ω1>ω2>ω3 and ω1, ω2 and ω3 are all greater than 0.

[0039] Furthermore, the generated health status assessment index is compared with the set health judgment threshold. Based on different comparison results, a corresponding health status judgment result is issued. The specific judgment logic is as follows:

[0040] When ZH≥1.0*yz′, the health status of the shrimp in the shrimp farming environment to be monitored is judged to be excellent, and the current environment is suitable for the growth and development of shrimp.

[0041] When 0.4*yz′≤ZH<1.0*yz′, the health status of the shrimp in the monitored shrimp farming environment is judged to be good, indicating that there is a risk of disease in the shrimp in the current environment, and corresponding measures should be taken to adjust it;

[0042] When 0≤ZHH<0.4*yz′, the health status of the shrimp in the shrimp farming environment to be monitored is judged to be poor, indicating that the shrimp in the current environment have developed diseases and should be reorganized.

[0043] Where yz′ is the health assessment threshold, which is obtained by correcting for the number of shrimp in the shrimp farming environment to be monitored. The formula used to calculate the health assessment threshold yz′ is as follows:

[0044]

[0045] In the formula, yz0 is the initial value of the health judgment threshold, DS is the number of shrimp in the shrimp farming environment to be monitored, and DS0 is the reference number.

[0046] Compared with the prior art, the beneficial effects of the present invention are:

[0047] First, this method precisely annotates shrimp images at different growth stages to obtain key image feature parameters, such as contour length, contour area, and shell contrast. The extraction of these features not only provides direct information about the shrimp's growth status but also lays the foundation for subsequent target recognition and feature extraction. By constructing target recognition and feature extraction models, automatic identification and analysis of individual shrimp can be achieved, improving work efficiency and reducing subjective errors from manual annotation. Second, by analyzing continuous time-series images of the aquaculture environment, this method can monitor shrimp behavioral characteristics in real time, such as swimming displacement and swimming frequency. This process not only improves the timeliness of shrimp health monitoring but also enhances the responsiveness to changes in the aquaculture environment. Combined with water source characteristic parameters (such as total bacterial count, pathogenic Vibrio count, and water quality indicators), the health status of shrimp can be comprehensively assessed. Finally, by comparing the generated health status assessment index with the set health judgment threshold, a health status judgment result can be quickly issued, forming an effective early warning mechanism. The ability to monitor and warn in real time helps farmers take timely measures to prevent disease occurrence and improve the success rate and economic benefits of shrimp farming. Attached Figure Description

[0048] Figure 1 This is a schematic diagram of the overall method flow of the present invention;

[0049] Figure 2A fitted curve of the shrimp average rate-behavior assessment influence coefficient;

[0050] Figure 3 The fitted curve of shrimp average swimming frequency versus behavioral assessment influence coefficient;

[0051] Figure 4 Calculate statistical curves for the influence coefficients of growth assessment;

[0052] Figure 5 This is a curve fitting the average contrast value to the growth assessment influence coefficient.

[0053] Figure 6 Statistical curves were calculated for the impact coefficient of disease incidence.

[0054] Figure 7 A bar chart for assessing the health status of shrimp;

[0055] Figure 8 This is a schematic diagram for identifying target standards for shrimp. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0057] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0058] Example:

[0059] Please see Figures 1-8 The present invention provides a technical solution:

[0060] A method for assessing shrimp health status based on multi-source information fusion, comprising the following steps:

[0061] Step 1: Acquire several shrimp images at different growth stages. Perform image preprocessing on the acquired shrimp images. Based on the preprocessed shrimp images, manually label each shrimp target in the preprocessed shrimp images using a bounding box. Obtain the image feature parameters corresponding to the labeled shrimp in the image, and record the labeled image as the training image. The image feature parameters include shrimp outline length, shrimp outline area, and shrimp shell contrast.

[0062] Several shrimp images at different growth stages are acquired, and the acquired shrimp images are preprocessed. The preprocessing includes image enhancement and denoising preprocessing. Wavelet transform is used to denoise each shrimp image, and bilateral filtering is used to enhance each shrimp image.

[0063] The method for denoising and enhancing the acquired shrimp images is as follows: The wavelet transform denoising method is used to denoise the distortion-corrected images. The specific steps of the wavelet transform denoising method include: decomposing the distortion-corrected image using wavelet transform to obtain wavelet coefficients at different scales and directions; thresholding the wavelet coefficients, setting low-amplitude wavelet coefficients to zero while retaining high-amplitude wavelet coefficients; and performing an inverse transform on the thresholded wavelet coefficients to reconstruct the image, thus completing the image denoising process.

[0064] Bilateral filtering was used to enhance the details of the acquired shrimp images. The specific formula used for the filtering transformation is as follows:

[0065]

[0066] In the formula, y is the coordinate vector in the image coordinate system, and I y Let B be the gray value at coordinate vector y. y Gray value I y The grayscale value G after bilateral filtering transformation d and G r Both are Gaussian functions, where G d and G r The formula used is:

[0067]

[0068] In the formula, x is the coordinate vector in the image coordinate system, and I x Let σ be the gray value at coordinate vector x. d and σ r G d and G r The standard deviation.

[0069] Each shrimp in the preprocessed shrimp image is manually labeled using bounding boxes. The specific method involves using an image annotation tool to label each shrimp in the image with a bounding box, where each bounding box is the smallest bounding rectangle of the shrimp. The LabelImg annotation tool is used for this purpose. After all annotations are completed, the results are saved in an appropriate format (such as XML, JSON, CSV, etc.) for subsequent model training and analysis.

[0070] The shrimp outline length refers to the total length of the object's boundary in the image, which is typically obtained using edge detection algorithms. Methods such as Canny edge detection or the Sobel operator are used to detect edges, followed by a contour detection function. Similarly, edge detection and contour detection methods are used to obtain the shrimp outline area.

[0071] Shrimp shell contrast can be obtained through the following steps: convert the image from the RGB color space to the HSV or Lab color space for better color analysis; use OpenCV functions to calculate a histogram; and then use it to calculate the color contrast. Specifically, shrimp shell contrast refers to the difference in grayscale value between each pixel and its neighboring pixels in the identified shrimp target image; the average of all pixel differences is taken as the shrimp shell contrast.

[0072] Step 2: Based on the obtained training images, establish a target recognition model and a feature extraction model. Use the training images as input to the target recognition model and the corresponding bounding boxes in the training images as labels to train the target recognition model. Crop the shrimp region image within the bounding boxes identified by the target recognition model and use it as input to the feature extraction model. Use the corresponding image feature parameters as labels to train the feature extraction model.

[0073] Based on the obtained training images, a target recognition model is established, which is based on the YOLOv5s network model. The trained target recognition model is used to identify and mark all shrimp targets in the shrimp image and extract the marked area image.

[0074] The primary function of the YOLOv5s network is real-time object detection. Object detection is a crucial task in computer vision, aiming to detect various categories of objects from images or videos and accurately label their locations. YOLOv5s, a model in the YOLO (You Only Look Once) family, strikes a balance between real-time performance and accuracy, making it suitable for scenarios requiring high speed and lightweight models.

[0075] A feature extraction model is established based on a convolutional neural network. This convolutional neural network consists of an input layer, convolutional layers, pooling layers, fully connected layers, and an output layer. The activation function in the convolutional layers is the ReLU function, and the specific expression of the ReLU function is as follows:

[0076] ReLU(s l(p,q) ) = max(0,s l(p,q) )

[0077] Where l represents the l-th corresponding convolutional layer, s l(p,q) Then it represents the q-th feature value of the p-th training sample image in the l-th corresponding convolutional layer, where l is the index of the convolutional layer, p is the index of the training sample image, and q is the index of the feature value in the training sample image;

[0078] For the fully connected layer, the number of neurons in the fully connected layer is set to 32, the initial neural network learning rate is set to 0.001, and the number of training rounds is 200.

[0079] The trained feature extraction model takes the labeled region image as input and outputs the corresponding predicted image feature parameters.

[0080] Step 3: Collect continuous time series images of the shrimp farming environment to be monitored within the collection period. After preprocessing the continuous time series images, input them into the trained target recognition model to identify the shrimp targets in the images and add bounding boxes to them. Input the image of the region within the bounding box into the trained feature extraction model to obtain the image feature parameters of the corresponding shrimp targets. Based on the generated bounding boxes and the image feature parameters of the corresponding shrimp targets, use the DeepSORT algorithm to extract the behavioral feature parameters of each shrimp target within the collection period.

[0081] The continuous time series images are preprocessed using methods consistent with those described above, which will not be repeated here. The specific logic for extracting behavioral feature parameters of each shrimp target within the acquisition time period using the DeepSORT algorithm is as follows: target detection is achieved by tracking and recognizing the bounding boxes of each shrimp output by the YOLOv5s network. After target detection, DeepSORT extracts feature representations for each detected shrimp. Feature extraction is performed using a convolutional neural network, and the extracted features are used for subsequent matching and tracking. DeepSORT uses a Kalman filter to predict and update the target state. Each target is modeled as a state variable, including its position, velocity, and other information. The Kalman filter can predict the next position in each frame based on the target's motion model, thus providing the target's trajectory. In each frame, DeepSORT performs data association (data association refers to matching detected targets with previously tracked targets) using the Hungarian algorithm based on the target detection results and the target position information from previous frames. The algorithm determines which newly detected targets are identical to which previously detected targets by calculating the distance between detected target features and existing target features. During data association, DeepSORT uses extracted features to determine the similarity between targets by calculating the Euclidean distance or cosine similarity between feature vectors. Once a target is successfully tracked, DeepSORT continuously updates the state of each shrimp and collects its behavioral feature parameters over time.

[0082] Step 4: Calculate and generate the behavior assessment influence coefficient and growth assessment influence coefficient based on the obtained behavioral and image feature parameters of the shrimp target. At the same time, obtain the water source feature parameters of the shrimp farming environment to be monitored during the collection period. Calculate the disease incidence influence coefficient based on the water source feature parameters. The behavioral feature parameters include shrimp swimming displacement and swimming frequency. The water source feature parameters include total bacterial count, number of pathogenic Vibrio bacteria, water temperature, and water pH.

[0083] The specific method for obtaining the water source characteristic parameters is as follows: Multiple sampling points are selected in the aquaculture water body, and water samples are collected using a sterile sampler. During sampling, care must be taken to avoid external contamination of the samples. The collected water samples are diluted, typically using a 10-fold dilution series. Then, the diluted water samples are inoculated onto culture media suitable for bacterial growth, such as nutrient agar (NA) or other selective media. The number of colonies cultured on each culture medium is counted using the plate count method (CFU), and the total number of colonies in the water sample is calculated.

[0084] Similarly, multiple sampling points are selected in the aquaculture water body, water samples are collected using a sterile sampler, and specific selective culture media, such as TCBS agar, which is suitable for the growth of Vibrio, are used to observe the colony characteristics on the culture medium, and the species and quantity of Vibrio are confirmed by biochemical experiments or molecular biological methods (such as PCR).

[0085] Use a water quality thermometer (such as a digital thermometer or water quality monitoring instrument) to directly measure the water temperature in the water body. Ensure that the thermometer probe is immersed in the water for a sufficient time to obtain an accurate reading.

[0086] Use a portable pH meter or a laboratory pH meter to take the measurement. Place the pH meter probe into the water sample and record the pH value after the instrument has stabilized.

[0087] Based on the obtained behavioral and image feature parameters of the shrimp target, behavioral assessment influence coefficients and growth assessment influence coefficients are calculated and generated. The specific formula for calculating the behavioral assessment influence coefficient is as follows:

[0088]

[0089] In the formula, CIF is the behavioral assessment impact coefficient, and V mean f represents the average swimming speed of all identified shrimp targets within the monitored shrimp farming environment during the data collection period. mean V0 represents the average number of swims of all identified shrimp targets within the monitored shrimp farming environment during the data collection period. V0 represents the set reference value for swimming speed, and f0 represents the set reference value for the number of swims. The average speed V0 of each swim during the data collection period is also considered. mean The calculation is performed based on the displacement and time of each swim, using the following formula:

[0090]

[0091] In the formula, V u Let be the average swimming speed of the u-th shrimp during each swimming period, where u is the index of the identified shrimp target, u∈[1,U], and U is the total number of identified shrimp targets.

[0092]

[0093] In the formula, X u,f Let Δt be the displacement of the u-th shrimp target during its f-th swim. u,f Let f be the time interval of the u-th shrimp target during the f-th swim, where f is the index of the number of swims of the u-th shrimp target during the collection period, and f∈[1,F].

[0094] It should be noted that the CIF (Contributing Factor for Behavioral Assessment) is used to assess the health status of shrimp through their behavioral parameters. The higher the CIF value, the better the health status of the shrimp.

[0095] The average number of times shrimp swam during the collection period, f mean This represents the activity frequency of shrimp over a certain period. When shrimp are subjected to external stimuli (such as deteriorating water quality, temperature changes, low oxygen levels, predator threats, etc.), they usually exhibit more frequent swimming. This is a direct response to environmental stress, meaning that the shrimp are trying to escape danger or cope with an unsuitable environment. Reduced swimming frequency often indicates that the shrimp may be sick or infected with parasites. In this case, the shrimp's vitality and function decline, and they exhibit a reluctance to move. Therefore, the average number of swimming frequency f of shrimp during the collection period is... mean The difference between the set reference value for the number of swims and the CIF (Contributing Factor to Behavioral Assessment) is inversely proportional to the CIF value, calculated using ln[1+|f]. mean The `-f0| / f0` parameter helps smooth out the impact of the number of swims, avoiding excessive amplification caused by too many swims. The logarithmic function's properties mean this value varies significantly within a small range, but gradually decreases at larger values, reflecting the diminishing marginal effect of the number of swims on health.

[0096] Within the shrimp farming environment to be monitored, the average swimming speed V of all identified shrimp targets during each swimming session within the data collection period. mean This represents the average swimming speed of shrimp. When shrimp are stimulated, such as by water pollution, temperature changes, low oxygen levels, or disease infection, they typically exhibit rapid swimming behavior. This rapid swimming is often a sign of stress, indicating that the shrimp is in an escape or defense state. Rapid swimming consumes a lot of energy, and prolonged stress and high energy consumption can lead to shrimp fatigue, affecting their growth and reproductive capacity, and even causing death. Slow swimming usually indicates that the shrimp may have health problems, such as bacterial, viral, or parasitic diseases, which can lead to reduced vitality. If the water quality is poor, shrimp may become listless and swim slowly due to lack of oxygen or toxins. Therefore, the average swimming speed V of shrimp during the sampling period is... mean The difference between the set swimming speed reference value and the behavior assessment influence coefficient (CIF) is inversely proportional to the value obtained through an exponential function. This reflects the significant impact of swimming speed on the evaluation coefficient. Table 1 shows the data for calculating the behavioral evaluation influence coefficient.

[0097] Table 1: Statistical Table for Calculation of Influence Coefficient of Behavioral Assessment

[0098]

[0099] Analysis of the data revealed a correlation between the behavioral evaluation influence coefficient and the average speed and average number of swims. For example, the data showed that the average speed fluctuated as the sample number increased; the average speed of sample number 7 was 7.0, significantly different from the 2.5 of sample number 11. This indicates that the average speed of the sample has a significant impact on the behavioral evaluation influence coefficient, especially at higher speeds, where the behavioral evaluation influence coefficient is generally higher.

[0100] When analyzing the relationship between the average number of swims and the influence coefficient of the behavioral assessment, a clear linear relationship was not observed between the average number of swims and the influence coefficient. For example, the average number of swims for sample number 3 was 15, and the influence coefficient of the behavioral assessment was only 0.33, while the average number of swims for sample number 10 was 14, and the influence coefficient was 2.84.

[0101] Among them, the average number of swimming times f of all identified shrimp targets within the monitored shrimp farming environment during the data collection period. mean The swimming speed reference value V0 is calculated by the number of times each shrimp swims during the collection period. The swimming speed reference value V0 is set according to the length of the collection period and the growth and development stage of the shrimp in the current shrimp farming environment, combined with expert experience. The swimming speed reference value f0 is set in the same way.

[0102] The specific formula used to calculate the growth assessment impact coefficient is as follows:

[0103]

[0104] In the formula, ECI is the growth assessment influence coefficient, and AC mean L represents the average shell contrast of all identified shrimp targets within the monitored shrimp farming environment. mean M represents the average length of the outlines of all shrimp targets identified within the shrimp farming environment to be monitored. mean L0 represents the average area of ​​all shrimp target outlines identified within the shrimp farming environment to be monitored, while M0 represents the reference values ​​for the shrimp target outline length and shrimp target outline area.

[0105] It should be noted that the Growth Assessment Impact Coefficient (ECI) assesses the health of shrimp by evaluating their growth status. The higher the ECI value, the healthier the shrimp are.

[0106] Among them, the outline length of shrimp is usually directly related to the shrimp's growth status. Shrimp with excessively large body length may have grown too quickly under certain conditions. This growth is usually uneven and may lead to weaker constitution and decreased resistance. Smaller shrimp may be affected by diseases, causing them to be unable to grow normally. Bacterial, viral, or parasitic infections often hinder the growth and development of shrimp. Therefore, the average outline length L of all identified target shrimp in the monitored shrimp farming environment is used as the basis for this study. mean The difference between the target outline length reference value and the growth assessment influence coefficient (ECI) is inversely proportional, and the outline area can be considered an approximate measure of shrimp volume. An excessively large outline area may indicate deformity, while a small outline area often indicates stunted growth and development. Therefore, the average M of the identified target outline areas of all shrimp in the monitored shrimp farming environment is used. mean The difference between the target outline area of ​​the shrimp and the reference value is inversely proportional to the growth assessment influence coefficient (ECI). This section combines information from both length and area dimensions, using the sum of squares to quantify the growth characteristics of shrimp. The sum of squares can be seen as a comprehensive measure of growth status in two-dimensional space, providing a more complete picture of shrimp growth.

[0107] The average length and area of ​​all identified shrimp target outlines within the monitored shrimp farming environment are obtained by averaging the predicted values ​​of all shrimp target outline lengths and areas output by the model. The reference values ​​for the shrimp target outline length and area are set based on the growth and development stage of the shrimp in the current monitored shrimp farming environment, combined with expert experience.

[0108] High shell contrast in shrimp may indicate abnormal pigmentation, disease, stress, or malnutrition. These health issues can alter the shrimp's appearance, affecting the color and texture of its shell. Therefore, high shell contrast (AC) in shrimp can indicate problems. for The growth assessment impact coefficient (ECI) is inversely proportional to the contrast ratio. An exponential function is used to ensure that higher contrast results in a lower ECI value, reflecting the significant negative impact of contrast on growth assessment. Partial data for calculating the growth assessment impact coefficient is shown in Table 2.

[0109] Table 2: Statistical Table of Growth Assessment Influence Coefficient Calculation

[0110]

[0111] Simultaneously, water source characteristic parameters of the shrimp farming environment to be monitored during the collection period are acquired, and the disease incidence impact coefficient is calculated based on the water source characteristic parameters. The specific formula used to calculate the disease incidence impact coefficient is as follows:

[0112]

[0113] In the formula, SCI is the disease incidence impact coefficient, RE is the total number of bacterial colonies in the water, HZ is the number of pathogenic Vibrio, and pH is... S The pH value of the water body, T S T0 represents the water temperature, PH0 represents the reference value for water pH, and T0 represents the reference value for water temperature.

[0114] It should be noted that the higher the SCI (Sickness Influence Coefficient) value, the worse the shrimp's growth environment, the more likely there are diseased shrimp, and therefore the worse the shrimp's health status.

[0115] The total bacterial count (RE) in water represents the concentration of bacteria in the water, especially the number of pathogens. A higher total bacterial count usually indicates poorer water hygiene and may increase the risk of disease outbreaks. Therefore, the total bacterial count (RE) is directly proportional to the disease incidence impact coefficient (SCI). The logarithmic function ln(1+RE) can smooth out the impact of small changes in bacterial count on the SCI, while avoiding numerical distortion caused by extremely high total bacterial counts.

[0116] Among them |PH S -PH0| 2 +|T S -T0| 2 These two items measure the deviation of the current pH and temperature of the water from the reference values. pH and temperature are important parameters of water quality. Deviation from the normal range will put stress on the health of shrimp and increase the possibility of disease. The square method is used to emphasize the impact of larger deviations, indicating that if the water parameters deviate from the reference values ​​significantly, the disease impact coefficient will increase significantly.

[0117] The number of pathogenic Vibrio bacteria, HZ, directly reflects the concentration of pathogenic Vibrio bacteria in the water. Pathogenic Vibrio bacteria are common pathogens in shrimp farming; therefore, the number of pathogenic Vibrio bacteria, HZ, is directly proportional to the disease incidence impact coefficient, SCI, through the exponential function e. -HZ The higher the number of pathogenic Vibrio bacteria in the denominator, the larger the SCI (Significance Influence Coefficient) for disease occurrence, reflecting their direct and significant threat to disease development. Data for the SCI calculation are shown in Table 3.

[0118] Table 3: Statistical Table for Calculation of Disease Incidence Impact Coefficient

[0119]

[0120] Step 5: Based on the obtained disease incidence impact coefficient, behavior assessment impact coefficient, and growth assessment impact coefficient, a health status assessment index is generated. The generated health status assessment index is compared with the set health judgment threshold. Based on different comparison results, the corresponding health status judgment result is issued to complete the shrimp health status assessment.

[0121] Based on the obtained disease incidence impact coefficient, behavioral assessment impact coefficient, and growth assessment impact coefficient, a health status assessment index is generated. The specific formula used to calculate the health status assessment index is as follows:

[0122]

[0123] In the formula, ZH is the health status assessment index, and ω1, ω2 and ω3 are the weight coefficients of the disease incidence influence coefficient, the behavioral assessment influence coefficient and the growth assessment influence coefficient, respectively, where ω1>ω2>ω3 and ω1, ω2 and ω3 are all greater than 0.

[0124] It should be noted that the Health Status Assessment Index ZH is used to represent the health status of shrimp. The higher the value, the better the health status of the shrimp.

[0125] Since the relative relationships between the various influence coefficients and the health status assessment index ZH have already been explained above, they will not be repeated here. The introduction of the logarithmic function ln(1+CIF) allows small changes in CIF to have a relatively small impact on the health status assessment index, while large changes in CIF will significantly affect the index calculation. This approach makes the health assessment less sensitive to minor behavioral changes, thus focusing more on overall behavioral trends. (Using square root form...) This reduces the impact of extreme values ​​and emphasizes smaller to moderate growth variations. It uses the square root form. Similarly.

[0126] Since disease outbreaks are often the primary cause of shrimp mortality and economic losses during aquaculture, the Disease Impact Coefficient (SCI) is assigned the highest weight. The Behavioral Assessment Impact Coefficient (CIF) is important because it reflects shrimp activity and appetite, but its impact on health status is smaller compared to disease risk. The Growth Assessment Impact Coefficient (ECI) reflects the shrimp's growth status; while its importance cannot be ignored, its impact is relatively small during acute disease outbreaks. Therefore, the parameters are set as follows: ω1 > ω2 > ω3, with ω1, ω2, and ω3 all greater than 0.

[0127] The generated health status assessment index is compared with the set health judgment threshold. Based on different comparison results, a corresponding health status judgment result is issued. The specific judgment logic is as follows:

[0128] When ZH≥1.0*yz′, the health status of the shrimp in the shrimp farming environment to be monitored is judged to be excellent, and the current environment is suitable for the growth and development of shrimp.

[0129] When 0.4*yz′≤ZH<1.0*yz′, the health status of the shrimp in the monitored shrimp farming environment is judged to be good, indicating that there is a risk of disease in the shrimp in the current environment, and corresponding measures should be taken to adjust it;

[0130] When 0 ≤ ZH < 0.4*yz′, the health status of the shrimp in the monitored shrimp farming environment is judged to be poor, indicating that the shrimp in the current environment have developed diseases and should be reorganized. Partial data on the calculation of the health status assessment index and the results of the health status judgment are shown in Table 4.

[0131] Table 4 Statistical Table of Health Status Assessment Results

[0132]

[0133]

[0134] Analysis of the data revealed a positive correlation between high health status assessment indices and good health status judgments in most samples. For example, sample number 4 had a health status assessment index of 0.3193, a behavioral assessment influence coefficient of 3.86, and a health status judgment of "excellent." This suggests that the sample performed well in the health status assessment, possibly due to its high growth assessment influence coefficient (0.4505) and relatively low disease incidence influence coefficient (15.4083), thereby improving its overall health status.

[0135] Furthermore, in analyzing the relationship between the incidence impact coefficient and the health status assessment index, we found that a higher incidence impact coefficient often corresponds to a lower health status assessment index. For example, sample number 3 had an incidence impact coefficient as high as 10.2342, while its health status assessment index was only 0.0653, resulting in a poor health status assessment. This indicates that an increase in the incidence impact coefficient may significantly reduce the health status of the sample.

[0136] Finally, combining the results of expert evaluations and health status assessments, a high degree of consistency between the expert evaluations and the assessment indices can be observed. For example, the health status assessment index for sample number 14 is 0.2197, with both the expert evaluation and the health status assessment indicating "excellent." This consistency demonstrates that this technical solution can effectively reflect the actual health status of the samples.

[0137] Where yz′ is the health assessment threshold, which is obtained by correcting for the number of shrimp in the shrimp farming environment to be monitored. The formula used to calculate the health assessment threshold yz′ is as follows:

[0138]

[0139] In the formula, yz0 is the initial value of the health judgment threshold, DS is the number of shrimp in the shrimp farming environment to be monitored, and DS0 is the reference number.

[0140] In shrimp farming, as the shrimp population increases, the bioburden in the water also increases, leading to changes in water quality (such as oxygen concentration and ammonia nitrogen concentration). These changes can put greater stress on shrimp health; higher shrimp densities result in greater competition for resources (such as food and habitat), increasing the risk of stress and disease transmission. Therefore, the shrimp population DS in the monitored shrimp farming environment is directly proportional to the health assessment threshold yz′.

[0141] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0142] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0143] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0144] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for assessing the health status of shrimp based on multi-source information fusion, characterized in that, The specific steps include: Several images of shrimp at different growth stages were acquired. The acquired shrimp images were preprocessed. Based on the preprocessed shrimp images, each shrimp target in the preprocessed shrimp images was labeled by a label box using a manual labeling method. The image feature parameters corresponding to the labeled shrimp in the image were obtained, and the labeled images were recorded as training images. The image feature parameters include shrimp outline length, shrimp outline area, and shrimp shell contrast. Based on the obtained training images, a target recognition model and a feature extraction model are established. The training images are used as input to the target recognition model, and the corresponding bounding boxes in the training images are used as labels to train the target recognition model. The shrimp region images identified by the target recognition model are cropped and used as input to the feature extraction model, and the corresponding image feature parameters are used as labels to train the feature extraction model. Continuous time-series images of the shrimp farming environment to be monitored were collected during the collection period. After preprocessing, the continuous time-series images were input into a trained target recognition model to identify shrimp targets in the images and add bounding boxes to them. The images of the regions within the bounding boxes were then input into a trained feature extraction model to obtain the image feature parameters corresponding to the shrimp targets. Based on the generated bounding boxes and the image feature parameters corresponding to the shrimp targets, the behavioral feature parameters of each shrimp target during the collection period were extracted using the DeepSORT algorithm. Based on the obtained behavioral and image feature parameters of the shrimp target, behavioral assessment influence coefficients and growth assessment influence coefficients are calculated and generated. At the same time, water source feature parameters of the shrimp farming environment to be monitored during the collection period are obtained, and disease incidence influence coefficients are calculated based on the water source feature parameters. The behavioral feature parameters include shrimp swimming displacement and swimming frequency, and the water source feature parameters include total bacterial count, number of pathogenic Vibrio bacteria, water temperature, and water pH value. Based on the obtained disease incidence impact coefficient, behavior assessment impact coefficient, and growth assessment impact coefficient, a health status assessment index is generated. The generated health status assessment index is compared with the set health judgment threshold. Based on different comparison results, the corresponding health status judgment result is issued to complete the shrimp health status assessment.

2. The method for assessing shrimp health status based on multi-source information fusion according to claim 1, characterized in that: Several shrimp images at different growth stages are acquired, and the acquired shrimp images are preprocessed. The preprocessing includes image enhancement and denoising preprocessing. Wavelet transform is used to denoise each shrimp image, and bilateral filtering is used to enhance each shrimp image. Each shrimp target in the preprocessed shrimp image is labeled using a manual labeling method. The specific labeling method is as follows: each shrimp target in the image is labeled using an image labeling tool, and the labeling box is the minimum bounding rectangle of each shrimp target.

3. The shrimp health status assessment method based on multi-source information fusion according to claim 2, characterized in that: Based on the obtained training images, a target recognition model is established, which is based on the YOLOv5s network model. The trained target recognition model is used to identify and mark all shrimp targets in the shrimp image and extract the marked area image. A feature extraction model is established based on a convolutional neural network. This convolutional neural network consists of an input layer, convolutional layers, pooling layers, fully connected layers, and an output layer. The activation function in the convolutional layers is the ReLU function, and the specific expression of the ReLU function is as follows: ReLU(s l(p,q) )=max(0,s l(p,q) ) Where l represents the l-th corresponding convolutional layer, s l(p,q) Then it represents the q-th feature value of the p-th training sample image in the l-th corresponding convolutional layer, where l is the index of the convolutional layer, p is the index of the training sample image, and q is the index of the feature value in the training sample image; For the fully connected layer, the number of neurons in the fully connected layer is set to 32, the initial neural network learning rate is set to 0.001, and the number of training rounds is 200. The trained feature extraction model takes the labeled region image as input and outputs the corresponding predicted image feature parameters.

4. The shrimp health status assessment method based on multi-source information fusion according to claim 1, characterized in that: Based on the obtained behavioral and image feature parameters of the shrimp target, behavioral assessment influence coefficients and growth assessment influence coefficients are calculated and generated. The specific formula for calculating the behavioral assessment influence coefficient is as follows: In the formula, CIF is the behavioral assessment impact coefficient, and V mean f represents the average swimming speed of all identified shrimp targets within the monitored shrimp farming environment during the data collection period. mean V0 represents the average number of swims of all identified shrimp targets within the monitored shrimp farming environment during the data collection period. V0 represents the set reference value for swimming speed, and f0 represents the set reference value for the number of swims. The average speed V0 of each swim during the data collection period is also represented by the value. mean The calculation is performed based on the displacement and time of each swim, using the following formula: In the formula, V u Let be the average swimming speed of the u-th shrimp during each swimming period, where u is the index of the identified shrimp target, u∈[1,U], and U is the total number of identified shrimp targets; In the formula, X u,f Let Δt be the displacement of the u-th shrimp target during its f-th swim. u,f Let f be the time interval of the u-th shrimp target during the f-th swim, where f is the index of the number of swims of the u-th shrimp target during the collection time period, F is the total number of swims of the u-th shrimp target during the collection time period, and f∈[1,F]. The specific formula used to calculate the growth assessment impact coefficient is as follows: In the formula, ECI is the growth assessment influence coefficient, and AC mean L represents the average shell contrast of all identified shrimp targets within the monitored shrimp farming environment. mean M represents the average length of the outlines of all shrimp targets identified within the shrimp farming environment to be monitored. mean L0 represents the average area of ​​all shrimp target outlines identified within the shrimp farming environment to be monitored, while M0 represents the reference values ​​for the shrimp target outline length and shrimp target outline area.

5. The shrimp health status assessment method based on multi-source information fusion according to claim 4, characterized in that: Simultaneously, water source characteristic parameters of the shrimp farming environment to be monitored during the collection period are acquired, and the disease incidence impact coefficient is calculated based on the water source characteristic parameters. The specific formula used to calculate the disease incidence impact coefficient is as follows: In the formula, SCI is the disease incidence impact coefficient, RE is the total number of bacterial colonies in the water, HZ is the number of pathogenic Vibrio, and pH is... S The pH value of the water body, T S T0 represents the water temperature, PH0 represents the reference value for water pH, and T0 represents the reference value for water temperature.

6. The shrimp health status assessment method based on multi-source information fusion according to claim 5, characterized in that: Based on the obtained disease incidence impact coefficient, behavioral assessment impact coefficient, and growth assessment impact coefficient, a health status assessment index is generated. The specific formula used to calculate the health status assessment index is as follows: In the formula, ZH is the health status assessment index, and ω1, ω2 and ω3 are the weight coefficients of the disease incidence influence coefficient, the behavioral assessment influence coefficient and the growth assessment influence coefficient, respectively, where ω1>ω2>ω3 and ω1, ω2 and ω3 are all greater than 0.

7. The shrimp health status assessment method based on multi-source information fusion according to claim 6, characterized in that: The generated health status assessment index is compared with the set health judgment threshold. Based on different comparison results, a corresponding health status judgment result is issued. The specific judgment logic is as follows: When ZH≥1.0*yz′, the health status of the shrimp in the shrimp farming environment to be monitored is judged to be excellent, and the current environment is suitable for the growth and development of shrimp. When 0.4*yz′≤ZH<1.0*yz′, the health status of the shrimp in the monitored shrimp farming environment is judged to be good, indicating that there is a risk of disease in the shrimp in the current environment, and corresponding measures should be taken to adjust it; When 0≤ZH<0.4*yz′, the health status of the shrimp in the shrimp farming environment to be monitored is judged to be poor, indicating that the shrimp in the current environment have developed diseases and should be reorganized. Where yz′ is the health assessment threshold, which is obtained by correcting for the number of shrimp in the shrimp farming environment to be monitored. The formula used to calculate the health assessment threshold yz′ is as follows: In the formula, yz0 is the initial value of the health judgment threshold, DS is the number of shrimp in the shrimp farming environment to be monitored, and DS0 is the reference number.

Images