Aquaculture organism health status monitoring system based on multidimensional feature association

Abstract

The present invention discloses a system for monitoring the health of aquaculture organisms based on multidimensional feature association, belonging to the field of aquaculture technology. The system comprises a visual feature acquisition layer, comprising an image acquisition module, an image preprocessing module, a target analysis module, and a feature extraction module; a molecular biology verification layer, comprising an immune gene monitoring module and a metabolic gene monitoring module; a data association prediction layer, comprising a standardization module, a feature screening module, a deep learning module, and an evaluation output module; and an adaptive optimization mechanism, a fault tolerance mechanism, and an environmental adaptability adjustment mechanism. The system provides a system for monitoring the health of aquaculture organisms based on multidimensional feature association, integrating computer vision technology and molecular biology detection methods to achieve intelligent monitoring of the health of aquaculture organisms, providing a scientific basis for aquaculture production management.

Description

Technical Field

[0001] The present invention relates to the technical field of aquaculture, and in particular to an aquaculture organism health status monitoring system based on multi-dimensional feature association. Background Art

[0002] With the large-scale and intensive development of aquaculture, monitoring the health status of farmed organisms has become a key link in ensuring aquaculture benefits and food safety. Existing aquaculture monitoring technologies mainly have the following problems and technical bottlenecks:

[0003] Traditional manual inspections rely heavily on the experience and judgment of livestock farmers. This approach is not only highly subjective and lacks standardization, but also lacks continuous monitoring due to limited human resources, making it easy to miss optimal opportunities for disease prevention and control. Furthermore, manual inspections lack quantifiable evaluation indicators, making it difficult to establish a systematic health status assessment system.

[0004] In recent years, computer vision-based monitoring systems have gained widespread application in aquaculture. These systems use video equipment to capture behavioral characteristics of aquacultured organisms and analyze them using image processing algorithms. However, existing visual monitoring systems have the following limitations: First, they require strict environmental conditions, significantly reducing monitoring effectiveness in low light conditions or turbid water. Second, current algorithms primarily focus on single behavioral characteristics and lack the ability to comprehensively analyze multidimensional behavioral features. Third, health status assessments based solely on visual features can be delayed, and some diseases may begin to develop before behavioral abnormalities become apparent.

[0005] Molecular biology detection technology can reflect the body's immune status and metabolic levels by detecting the expression levels of specific genes, with high diagnostic accuracy. Currently, there are a variety of PCR-based detection solutions on the market, but these methods generally suffer from long detection cycles and high costs. More importantly, due to the lack of correlation analysis with real-time behavioral monitoring data, it is difficult to establish a correspondence between changes at the molecular level and changes in apparent behavior, resulting in the inability to provide timely warnings and predict the development of disease trends.

[0006] Existing research on the application of deep learning technology primarily focuses on feature extraction and classification from a single data source, lacking the ability to integrate and analyze heterogeneous data from multiple sources. While some studies have attempted to combine visual features with environmental parameters, no systematic approach has yet achieved a deep correlation between visual features and molecular markers, significantly limiting the accuracy and predictive power of health status assessments.

[0007] Existing monitoring systems generally lack adaptive optimization mechanisms. In real-world aquaculture environments, due to the dynamic changes in factors such as water quality, temperature, and light, fixed monitoring parameters and evaluation criteria often cannot adapt to these environmental changes, compromising the reliability of monitoring results. Furthermore, the system's fault tolerance and recovery mechanisms are relatively weak, making it difficult to maintain basic monitoring functions in the event of equipment failures or data transmission anomalies.

[0008] Therefore, how to organically combine real-time visual monitoring with periodic molecular biology verification to build an intelligent monitoring system with adaptive optimization capabilities and achieve accurate assessment and early warning of the health status of farmed organisms is a technical problem that needs to be urgently solved in the current aquaculture field. Summary of the Invention

[0009] The purpose of the present invention is to provide an aquaculture organism health status monitoring system based on multi-dimensional feature association to solve the problems existing in the background technology.

[0010] To achieve the above objectives, the present invention provides an aquaculture organism health status monitoring system based on multi-dimensional feature association, comprising:

[0011] The visual feature acquisition layer, consisting of an image acquisition module, an image preprocessing module, a target analysis module, and a feature extraction module, enables continuous monitoring and analysis of fish swimming speed, clustering status, and gill cover movement behavior characteristics;

[0012] The molecular biology validation layer, consisting of an immune gene monitoring module and a metabolic gene monitoring module, is used to detect the expression levels of specific genes, including immune-related genes IL-1β and TNF-α, and metabolic-related genes GLUT and FAS. After standardization, the test data is stored in a dedicated database, and a baseline value and fluctuation range system for gene expression levels is established based on this data.

[0013] The data association prediction layer, consisting of a normalization module, a feature screening module, a deep learning module, and an evaluation output module, is used to achieve rapid prediction of health status from real-time visual features to the molecular level, and to update and optimize the prediction model online through regular molecular biology verification data;

[0014] The adaptive optimization mechanism can cross-validate the behavioral feature data obtained by the visual feature acquisition layer and the gene expression data obtained by the molecular biology verification layer. It automatically triggers the update of model parameters when the consistency between the predicted results and the actual gene expression levels is less than 85%, and automatically adjusts the sampling frequency and scoring threshold of visual features according to changes in environmental parameters. At the same time, it maintains the data cache of the last 24 hours and has automatic recovery capabilities in the event of data transmission interruption or equipment failure, thereby realizing intelligent monitoring and early warning of the health status of aquaculture organisms.

[0015] Preferably, the contents of the image acquisition module, image preprocessing module, target analysis module and feature extraction module are as follows:

[0016] Image acquisition module, including a high-definition camera with a frame rate of 30fps, used to collect video stream data of the aquaculture pond with a resolution of 1920×1080;

[0017] The image preprocessing module is used to perform inter-frame difference calculation and adaptive binarization processing on the collected video stream to extract the moving target in the video;

[0018] The target analysis module uses the YOLO v5 detector to detect fish targets and count their number in the preprocessed images;

[0019] The feature extraction module is used to calculate the swimming speed and trajectory changes per unit time by using the target displacement of the previous and next frames. It analyzes the spatial distribution characteristics and clustering status of the fish school based on the nearest neighbor distance algorithm, and uses the local sensitive hashing technology to match the gill cover area to realize the monitoring of the gill cover movement frequency.

[0020] The feature extraction module performs feature calculation based on the fish body detection results obtained by the target analysis module, improves the accuracy of feature extraction through the moving target information provided by the image preprocessing module, and realizes real-time monitoring of fish behavior characteristics.

[0021] Preferably, the contents of the immune gene monitoring module and the metabolic gene monitoring module are as follows:

[0022] Immune gene monitoring module, used to detect the expression levels of interleukin IL-1β and tumor necrosis factor TNF-α, to achieve molecular verification of the immune response status of fish;

[0023] Metabolic gene monitoring module, used to detect the expression levels of glucose transporter GLUT and fatty acid synthase FAS, to achieve molecular verification of fish metabolic activity;

[0024] The molecular biology verification layer establishes a molecular characteristic map of the sub-health status of fish through a combined analysis of the above-mentioned gene expression levels, providing a reliable biological verification basis for the data association prediction layer.

[0025] Preferably, the immune gene monitoring module includes an inflammatory response analysis unit and an immune function assessment unit;

[0026] The inflammatory response analysis unit detects the gene expression patterns of IL-1β and TNF-α. When the IL-1β gene expression level increases to more than 4 times the basal level and the TNF-α gene expression level increases to more than 3 times the basal level, it is determined to be an acute inflammatory response state. When the IL-1β gene expression level increases to 2-4 times the basal level and the TNF-α gene expression level increases to 1.5-3 times the basal level, it is determined to be a chronic inflammatory response state.

[0027] The immune function assessment unit conducts a comprehensive assessment of the activation level and functional status of the immune system based on the expression combination characteristics of inflammatory factor genes.

[0028] Preferably, the metabolic gene monitoring module includes a metabolic activity assessment unit and an energy balance analysis unit;

[0029] The metabolic activity assessment unit monitors the expression changes of GLUT and FAS genes. When the GLUT gene expression level drops to less than 0.5 times the basal level and the FAS gene expression level drops to less than 0.3 times the basal level, it is judged as a severe metabolic inhibition state. When the GLUT gene expression level drops to 0.5-0.8 times the basal level and the FAS gene expression level drops to 0.3-0.7 times the basal level, it is judged as a moderate metabolic inhibition state.

[0030] The energy balance analysis unit combines the expression characteristics of metabolism-related genes to dynamically monitor the body's energy metabolism level and nutritional status.

[0031] Preferably, the contents of the standardization module, feature screening module, deep learning module and evaluation output module are as follows:

[0032] The standardization module is used to perform Min-Max standardization and time-series alignment on the multi-dimensional data input from the visual feature acquisition layer and the molecular biology verification layer, achieving unified expression of heterogeneous data;

[0033] The feature screening module is used to optimize and reduce the dimension of standardized features based on data correlation and information entropy theory, and dynamically sample time series features through a sliding window;

[0034] The deep learning module uses an LSTM network structure to perform temporal modeling on the filtered features and implements correlation learning and pattern mining between visual features and molecular markers through a multi-layer attention mechanism;

[0035] The evaluation output module is used to generate a health status score from 0 to 100 points and classify the scores based on preset thresholds to achieve a visual early warning of the health status of the farmed organisms;

[0036] The data association prediction layer uses a mapping model trained by a deep learning module to achieve rapid prediction from real-time visual features to molecular-level health status, and updates and optimizes the prediction model online through regular molecular biology verification data.

[0037] Preferably, the intelligent monitoring process of the health status of aquaculture organisms by the data association prediction layer includes the following steps:

[0038] A 24-hour sliding window was used to segment the visual feature data into time series, and the swimming speed, clustering state, and gill cover movement characteristics were normalized to the range of 0-1 using the minimum-maximum normalization method.

[0039] A feature vector was constructed based on the mean and variance of the visual features in each time window. A training sample set was established by combining the gene expression data at the corresponding time point. The dataset was divided into a training set and a validation set in an 8:2 ratio using a stratified sampling method.

[0040] The multi-head self-attention mechanism is used to perform temporal modeling of visual feature sequences. This mechanism can divide the visual features at each moment into multiple subspaces and independently calculate weights in each subspace to obtain the output of each head. The outputs of each head are then concatenated together and a linear transformation is used to achieve feature representation after temporal modeling. The encoded feature vector is then deeply integrated with gene expression data, and cross-validation is used to select the optimal model parameters.

[0041] Based on the model output, a health status scoring system with a score of 0-100 is established, where 0-60 is abnormal, 61-75 is sub-healthy, 76-90 is good, and 91-100 is optimal.

[0042] The data association prediction layer uses newly added molecular biology verification data to update the model online every 48 hours, evaluates the model performance by calculating the consistency between the predicted results and the actual gene expression levels, and dynamically adjusts the warning threshold based on the verification results.

[0043] Preferably, a fault tolerance mechanism is also included:

[0044] It includes data acquisition exception processing unit, equipment fault detection unit and system recovery unit;

[0045] When the data collection exception processing unit detects that the video stream data collection is interrupted, it automatically switches to the backup camera to continue data collection, and records the time when the exception occurred and the duration;

[0046] The equipment fault detection unit monitors the operating status and data transmission quality of each functional module. When a device fault or data transmission anomaly is detected, it automatically switches the relevant module to safe mode and issues an alarm message.

[0047] The system recovery unit is responsible for data recovery and system parameter recalibration after troubleshooting. It conducts a retrospective assessment of the health status during the interruption by calling historical data, and performs an additional molecular biology verification within 72 hours after resuming normal operation to ensure the continuity and reliability of the system monitoring results.

[0048] Preferably, it also includes an environmental adaptability adjustment mechanism, which automatically adjusts the exposure parameters of the camera according to the light intensity, and starts the infrared fill light mode when the light is lower than 100 lux; adjusts the threshold of the image preprocessing algorithm based on water quality parameters, and adds an image enhancement processing step when the water turbidity exceeds 50NTU; automatically corrects the evaluation criteria of behavioral characteristics according to water temperature changes, and adjusts the baseline values of swimming speed and gill cover movement frequency accordingly when the water temperature is lower than 15℃ or higher than 30℃; automatically updates the calculation parameters of group behavioral characteristics when the breeding density changes significantly, and triggers a molecular biological verification when the density changes by more than 30%, so as to ensure the monitoring accuracy of the system under different environmental conditions.

[0049] Preferably, the visual feature acquisition layer monitors the fish behavior characteristics including the following corresponding relationships:

[0050] Swimming speed characteristics: By monitoring the displacement changes of fish per unit time, when the swimming speed is less than 50% of the normal value or the swimming trajectory shows irregular jitter, it indicates that the fish may be in a state of stress or abnormal energy metabolism. When the group swimming speed suddenly increases by more than 200% of the normal value, it indicates that the water quality may be deteriorating or there may be external interference.

[0051] Clustering characteristics are determined by calculating the spatial distance distribution between individuals. When the nearest neighbor distance of more than 90% of the individuals is less than 30% of the normal value, it indicates that the fish are in a state of stress or fright. When the degree of individual dispersion increases significantly and the nearest neighbor distance is greater than 300% of the normal value, it indicates that there may be infectious diseases or insufficient dissolved oxygen in the water.

[0052] The gill cover movement characteristics are measured by measuring the frequency of gill cover opening and closing. When the gill cover movement frequency exceeds 150% of the normal value and this state lasts for more than 2 hours, it indicates that the fish is in a state of respiratory distress or dissolved oxygen deficiency. When the gill cover movement frequency of more than 50% of the individuals is less than 60% of the normal value, it indicates decreased metabolic activity or environmental stress.

[0053] The visual feature acquisition layer establishes a mapping relationship between the fish's apparent behavior and health status through combined analysis of the above behavioral features, providing real-time behavioral basis for the data association prediction layer.

[0054] Therefore, the present invention adopts the above-mentioned aquaculture organism health status monitoring system based on multidimensional feature association, which has the following beneficial effects:

[0055] (1) The system achieves a deep fusion of visual features and molecular markers, and significantly improves the accuracy and reliability of health status monitoring by establishing a mapping relationship between behavioral phenotypes and gene expression levels. The system adopts an LSTM network structure and a multi-layer attention mechanism, which can effectively capture the temporal correlation and causal relationship between different features, providing a more comprehensive scientific basis for health status assessment.

[0056] (2) A monitoring system combining real-time visual monitoring with periodic molecular verification has been established, which not only ensures the real-time nature of monitoring but also the reliability of the evaluation results. Computer vision technology is used to continuously monitor the behavioral characteristics of aquaculture organisms, and the prediction model is optimized and updated in combination with regular molecular biological verification data, which significantly reduces the detection cost while ensuring the monitoring effect.

[0057] (3) Deep learning methods are used to achieve correlation analysis of multi-dimensional features. By learning from a large amount of historical data, a prediction model with strong adaptability and good generalization ability is established. The system can quickly assess the health status of farmed organisms based on real-time visual features, and continuously optimize model parameters through regular molecular biological verification to improve prediction accuracy;

[0058] (4) A graded early warning mechanism was designed, which quantified the health status into a scoring system of 0-100 points and set corresponding early warning thresholds. This visual evaluation method makes it easier for aquaculture managers to intuitively understand the monitoring results and take appropriate preventive and treatment measures in a timely manner, thereby improving the efficiency and scientific nature of aquaculture management.

[0059] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 Schematic diagram of the overall architecture of the system of the present invention;

[0061] Figure 2 This is a functional module structure diagram of the visual feature acquisition layer of the present invention;

[0062] Figure 3 This is the workflow diagram for the molecular biology validation layer;

[0063] Figure 4 Schematic diagram of the network structure of the data association prediction layer. DETAILED DESCRIPTION

[0064] The following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but rather merely represents selected embodiments of the present invention. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without inventive effort shall fall within the scope of protection of the present invention.

[0065] See also Figure 1 , a system for monitoring the health status of aquaculture organisms based on multidimensional feature association, comprising:

[0066] The visual feature acquisition layer, consisting of an image acquisition module, an image preprocessing module, a target analysis module, and a feature extraction module, enables continuous monitoring and analysis of behavioral characteristics such as fish swimming speed, clustering status, and gill cover movement.

[0067] The image acquisition module includes a high-definition camera with a frame rate of 30fps, which is used to collect 1920×1080 resolution video stream data of the aquaculture pond; and uses a built-in adaptive light compensation algorithm to maintain stable image acquisition quality within the lighting range of 20-10,000 lux.

[0068] The image preprocessing module is used to perform inter-frame difference operation and adaptive binarization processing on the collected video stream to extract the moving target in the video.

[0069] The target analysis module uses the YOLO v5 detector to perform fish target detection and quantity counting on the preprocessed images.

[0070] The feature extraction module is used to calculate the swimming speed and trajectory changes per unit time through the target displacement of the previous and next frames, analyze the spatial distribution characteristics and clustering status of the fish school based on the nearest neighbor distance algorithm, and use the local sensitive hashing technology to match the gill cover area to realize the monitoring of the gill cover movement frequency.

[0071] The feature extraction module performs feature calculation based on the fish body detection results obtained by the target analysis module, improves the accuracy of feature extraction through the moving target information provided by the image preprocessing module, and realizes real-time monitoring of fish behavior characteristics.

[0072] The visual feature acquisition layer monitors fish behavior characteristics, including the following corresponding relationships:

[0073] The swimming speed characteristics are determined by monitoring the displacement changes of fish per unit time. When the swimming speed is lower than 50% of the normal value or the swimming trajectory shows irregular jitters, it indicates that the fish may be in a state of stress or have abnormal energy metabolism. When the swimming speed of the group suddenly increases by more than 200% of the normal value, it indicates that there may be water quality deterioration or external interference.

[0074] Clustering status characteristics are determined by calculating the spatial distance distribution between individuals. When the nearest neighbor distance of more than 90% of the individuals is less than 30% of the normal value, it indicates that the fish are in a state of stress or fright; when the degree of individual dispersion increases significantly and the nearest neighbor distance is greater than 300% of the normal value, it indicates the possible presence of infectious diseases or insufficient dissolved oxygen in the water.

[0075] The gill cover movement characteristics are measured by measuring the frequency of gill cover opening and closing. When the gill cover movement frequency exceeds 150% of the normal value and this state lasts for more than 2 hours, it indicates that the fish is in a state of difficulty breathing or insufficient dissolved oxygen; when the gill cover movement frequency of more than 50% of individuals is lower than 60% of the normal value, it indicates a decrease in metabolic activity or environmental stress.

[0076] The visual feature acquisition layer establishes a mapping relationship between the fish's apparent behavior and health status through combined analysis of the above behavioral features, providing real-time behavioral basis for the data association prediction layer.

[0077] The molecular biology verification layer consists of an immune gene monitoring module and a metabolic gene monitoring module. It uses fluorescent quantitative PCR technology to detect the expression levels of specific genes, including immune-related genes IL-1β and TNF-α and metabolic-related genes GLUT and FAS, on a 48-hour cycle. After standardization, the test data is stored in an exclusive database, and on this basis, a baseline value and fluctuation range system for gene expression levels is established.

[0078] The immune gene monitoring module is used to detect the expression levels of interleukin IL-1β and tumor necrosis factor TNF-α, thereby achieving molecular verification of the immune response status of fish.

[0079] The metabolic gene monitoring module is used to detect the expression levels of glucose transporter GLUT and fatty acid synthase FAS, realizing molecular verification of fish metabolic activity.

[0080] The molecular biology verification layer establishes a molecular characteristic map of the sub-health status of fish through a combined analysis of the above-mentioned gene expression levels, providing a reliable biological verification basis for the data association prediction layer.

[0081] The immune gene monitoring module includes an inflammatory response analysis unit and an immune function assessment unit;

[0082] The inflammatory response analysis unit detects the gene expression patterns of IL-1β and TNF-α. When the IL-1β gene expression level increases to more than 4 times the basal level and the TNF-α gene expression level increases to more than 3 times the basal level, it is determined to be an acute inflammatory response state. When the IL-1β gene expression level increases to 2-4 times the basal level and the TNF-α gene expression level increases to 1.5-3 times the basal level, it is determined to be a chronic inflammatory response state.

[0083] The immune function assessment unit conducts a comprehensive assessment of the activation level and functional status of the immune system based on the expression combination characteristics of inflammatory factor genes.

[0084] The metabolic gene monitoring module includes a metabolic activity assessment unit and an energy balance analysis unit;

[0085] The metabolic activity assessment unit monitors the expression changes of GLUT and FAS genes. When the GLUT gene expression level drops to less than 0.5 times the basal level and the FAS gene expression level drops to less than 0.3 times the basal level, it is judged as a severe metabolic inhibition state. When the GLUT gene expression level drops to 0.5-0.8 times the basal level and the FAS gene expression level drops to 0.3-0.7 times the basal level, it is judged as a moderate metabolic inhibition state.

[0086] The energy balance analysis unit combines the expression characteristics of metabolism-related genes to dynamically monitor the body's energy metabolism level and nutritional status.

[0087] The data association prediction layer, consisting of a standardization module, a feature screening module, a deep learning module, and an evaluation output module, is used to achieve rapid prediction from real-time visual features to molecular-level health status, and to update and optimize the prediction model online through regular molecular biology verification data.

[0088] The standardization module is used to perform Min-Max standardization and time series alignment on the multi-dimensional data input by the visual feature acquisition layer and the molecular biology verification layer to achieve unified expression of heterogeneous data.

[0089] The feature screening module is used to optimize and reduce the dimension of standardized features based on data correlation and information entropy theory, and dynamically sample time series features through a sliding window.

[0090] The deep learning module uses an LSTM network structure to perform temporal modeling on the filtered features. A multi-layer attention mechanism is used to learn and mine patterns in the associations between visual features and molecular markers. A deep learning model with a three-layer LSTM network structure and 128 neurons per layer is used to model temporal features. A four-layer multi-head attention mechanism is used to analyze the associations between features, with the number of attention heads set to 8.

[0091] The evaluation output module is used to generate a health status score of 0-100 points and grade the scores based on preset thresholds to achieve visual early warning of the health status of farmed organisms.

[0092] The data association prediction layer uses a mapping model trained by a deep learning module to achieve rapid prediction from real-time visual features to molecular-level health status, and updates and optimizes the prediction model online through regular molecular biology verification data.

[0093] The intelligent monitoring process of the health status of aquaculture organisms by the data association prediction layer includes the following steps:

[0094] A 24-hour sliding window was used to segment the visual feature data into time series, and the swimming speed, clustering state, and gill cover movement characteristics were normalized to the range of 0-1 using the minimum-maximum normalization method.

[0095] A feature vector was constructed based on the mean and variance of the visual features in each time window. A training sample set was established by combining the gene expression data at the corresponding time point. The dataset was divided into a training set and a validation set in an 8:2 ratio using a stratified sampling method.

[0096] The multi-head self-attention mechanism is used to perform temporal modeling of visual feature sequences. The multi-head self-attention mechanism can divide the visual features at each moment into multiple subspaces, and independently calculate the weights in each subspace to obtain the output of each head. The outputs of each head are spliced together, and the feature representation after temporal modeling is realized through a linear transformation. The encoded feature vector is then deeply fused with the gene expression data, and the optimal model parameters are selected using the cross-validation method. In other words, the multi-head self-attention mechanism can perform multi-dimensional weight analysis on each moment, and then sort the weight analysis results at different moments according to the time series, and then deeply fuse the encoded feature vector with the gene expression data.

[0097] Based on the model output, a health status scoring system with a score of 0-100 is established, where 0-60 is abnormal, 61-75 is sub-healthy, 76-90 is good, and 91-100 is optimal.

[0098] The data association prediction layer uses newly added molecular biology verification data to update the model online every 48 hours, evaluates the model performance by calculating the consistency between the predicted results and the actual gene expression levels, and dynamically adjusts the warning threshold based on the verification results.

[0099] Each functional layer is interconnected through a data bus to jointly monitor the health status of farmed organisms.

[0100] The adaptive optimization mechanism can cross-validate the behavioral feature data obtained by the visual feature acquisition layer and the gene expression data obtained by the molecular biology verification layer. It automatically triggers the update of model parameters when the consistency between the predicted results and the actual gene expression levels is less than 85%, and automatically adjusts the sampling frequency and scoring threshold of visual features according to changes in environmental parameters. At the same time, it maintains the data cache of the last 24 hours and has automatic recovery capabilities in the event of data transmission interruption or equipment failure, thereby realizing intelligent monitoring and early warning of the health status of aquaculture organisms.

[0101] Also includes fault tolerance mechanisms:

[0102] It includes data acquisition exception processing unit, equipment fault detection unit and system recovery unit;

[0103] When the data collection exception processing unit detects that the video stream data collection is interrupted, it automatically switches to the backup camera to continue data collection, and records the time when the exception occurred and the duration;

[0104] The equipment fault detection unit monitors the operating status and data transmission quality of each functional module. When a device fault or data transmission anomaly is detected, it automatically switches the relevant module to safe mode and issues an alarm message.

[0105] The system recovery unit is responsible for data recovery and system parameter recalibration after troubleshooting. It conducts a retrospective assessment of the health status during the interruption by calling historical data, and performs an additional molecular biology verification within 72 hours after resuming normal operation to ensure the continuity and reliability of the system monitoring results.

[0106] It also includes an environmental adaptability adjustment mechanism, which automatically adjusts the camera's exposure parameters according to the light intensity and activates the infrared fill light mode when the light is lower than 100 lux; adjusts the threshold of the image preprocessing algorithm based on water quality parameters, and adds an image enhancement processing step when the water turbidity exceeds 50NTU; automatically corrects the evaluation criteria of behavioral characteristics according to changes in water temperature, and adjusts the baseline values of swimming speed and gill cover movement frequency accordingly when the water temperature is lower than 15℃ or higher than 30℃; automatically updates the calculation parameters of group behavioral characteristics when the aquaculture density changes significantly, and triggers a molecular biological verification when the density changes by more than 30%, to ensure the monitoring accuracy of the system under different environmental conditions.

[0107] The system uses the data association prediction layer to analyze the behavioral feature data obtained by the visual feature acquisition layer and the gene expression data obtained by the molecular biology verification layer, thereby combining high-precision molecular detection with real-time visual monitoring to conduct intelligent monitoring of the health status of aquaculture organisms.

[0108] The monitoring system is deployed in a standardized breeding pond at a breeding base. Figure 1 The system architecture shown here features a camera array installed above the aquaculture ponds to collect data for the visual feature acquisition layer. A molecular biology laboratory is established within the farm for gene expression testing. A data processing server runs the deep learning model for the data association prediction layer. Data transmission between these functional layers is achieved via a local area network, ensuring the coordinated operation of all system components.

[0109] like Figure 2 As shown in FIG, the structure of the visual feature acquisition layer includes an image acquisition module, an image preprocessing module, a target analysis module and a feature extraction module.

[0110] During actual monitoring, the image acquisition module captures water surface images using a high-definition camera mounted above the aquaculture pond. The image preprocessing module performs inter-frame differencing on the raw video stream to extract moving targets. The target analysis module uses the YOLOv5 detector to identify fish in the video. The feature extraction module calculates characteristic parameters such as swimming speed, group distribution, and gill cover movement. The cascaded processing of these modules ensures accurate acquisition of visual feature data.

[0111] like Figure 3 As shown in the figure, the workflow of the molecular biology verification layer mainly includes two parts: immune gene monitoring and metabolic gene monitoring.

[0112] During the breeding process, tissue samples are regularly collected from the cultured organisms for gene expression testing. The immune gene monitoring module uses real-time fluorescence quantitative PCR to detect the expression levels of IL-1β and TNF-α, while the metabolic gene monitoring module detects the expression levels of GLUT and FAS. After data normalization, the test results are stored in the system database for subsequent model validation and updates.

[0113] like Figure 4 As shown in the figure, the data association prediction layer adopts a deep learning network structure, including modules such as data normalization, feature screening, LSTM time series modeling, and multi-layer attention mechanism.

[0114] During the model training and application phase, the collected visual feature data and molecular biology data are first standardized. The standardized feature data is then processed by the feature screening module for dimensionality reduction and then fed into the LSTM network for time series modeling. The network's output is processed by the evaluation module to generate a health status score.

[0115] As shown in Table 1, the system adopts a hierarchical warning mechanism to divide health status scores into different levels.

[0116] During daily monitoring, the system calculates health status scores in real time and issues warnings based on the grading criteria shown in Table 1. When the score falls below the warning threshold, the system automatically sends an alert to livestock managers, prompting the need for appropriate management intervention. The system also records all warning events and their corresponding characteristic data for continuous optimization of the prediction model.

[0117] Table 1 Health status score and warning level classification

[0118] ;

[0119] Therefore, the present invention adopts the above-mentioned aquaculture organism health status monitoring system based on multidimensional feature association, and realizes intelligent monitoring of the health status of aquaculture organisms by integrating computer vision technology and molecular biology detection methods. It has strong adaptability and can be promoted and applied to health monitoring of different types of aquaculture organisms by adjusting feature extraction strategies and verification indicators.

[0120] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the same. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solutions of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A system for monitoring the health status of aquaculture organisms based on multidimensional feature association, characterized in that: include: The visual feature acquisition layer, consisting of an image acquisition module, an image preprocessing module, a target analysis module, and a feature extraction module, enables continuous monitoring and analysis of fish swimming speed, clustering status, and gill cover movement behavior characteristics; The molecular biology validation layer, consisting of an immune gene monitoring module and a metabolic gene monitoring module, is used to detect the expression levels of specific genes, including immune-related genes IL-1β and TNF-α, and metabolic-related genes GLUT and FAS. After standardization, the test data is stored in a database, and a baseline value and fluctuation range system for gene expression levels is established. The data association prediction layer, consisting of a normalization module, a feature screening module, a deep learning module, and an evaluation output module, is used to achieve rapid prediction of health status from real-time visual features to the molecular level, and to update and optimize the prediction model online through regular molecular biology verification data; The intelligent monitoring process of the health status of aquaculture organisms by the data association prediction layer includes the following steps: A 24-hour sliding window was used to segment the visual feature data into time series, and the swimming speed, clustering state, and gill cover movement characteristics were normalized to the range of 0-1 using the minimum-maximum normalization method. A feature vector was constructed based on the mean and variance of the visual features in each time window. A training sample set was established by combining the gene expression data at the corresponding time point. The dataset was divided into a training set and a validation set in an 8:2 ratio using a stratified sampling method. The multi-head self-attention mechanism is used to perform temporal modeling of visual feature sequences. This mechanism can divide the visual features at each moment into multiple subspaces and independently calculate weights in each subspace to obtain the output of each head. The outputs of each head are then concatenated together and a linear transformation is used to achieve feature representation after temporal modeling. The encoded feature vector is then deeply integrated with gene expression data, and cross-validation is used to select the optimal model parameters. Based on the model output, a health status scoring system with a score of 0-100 is established, where 0-60 is abnormal, 61-75 is sub-healthy, 76-90 is good, and 91-100 is optimal. The adaptive optimization mechanism performs cross-validation based on the behavioral feature data obtained by the visual feature acquisition layer and the gene expression data obtained by the molecular biology verification layer. It automatically triggers model parameter updates when the consistency between the predicted results and the actual gene expression levels is less than 85%, and automatically adjusts the sampling frequency and scoring threshold of visual features according to changes in environmental parameters. At the same time, it maintains the data cache of the last 24 hours and has automatic recovery capabilities in the event of data transmission interruption or equipment failure.

2. The aquaculture organism health status monitoring system based on multidimensional feature association according to claim 1 is characterized in that: The contents of the image acquisition module, image preprocessing module, target analysis module and feature extraction module are as follows: Image acquisition module, including a high-definition camera with a frame rate of 30fps, used to collect video stream data of the aquaculture pond with a resolution of 1920×1080; The image preprocessing module is used to perform inter-frame difference calculation and adaptive binarization processing on the collected video stream to extract the moving target in the video; The target analysis module uses the YOLO v5 detector to detect fish targets and count their number in the preprocessed images; The feature extraction module is used to calculate the swimming speed and trajectory changes per unit time through the target displacement of the previous and next frames, analyze the spatial distribution characteristics and clustering status of the fish school based on the nearest neighbor distance algorithm, and use the local sensitive hashing technology to match the gill cover area to realize the monitoring of the gill cover movement frequency.

3. The aquaculture organism health status monitoring system based on multidimensional feature association according to claim 2 is characterized in that: The contents of the immune gene monitoring module and metabolic gene monitoring module are as follows: Immune gene monitoring module, used to detect the expression levels of interleukin IL-1β and tumor necrosis factor TNF-α, to achieve molecular verification of the immune response status of fish; The metabolic gene monitoring module is used to detect the expression levels of glucose transporter GLUT and fatty acid synthase FAS, realizing molecular verification of fish metabolic activity.

4. The aquaculture organism health status monitoring system based on multidimensional feature association according to claim 3 is characterized by: The immune gene monitoring module includes an inflammatory response analysis unit and an immune function assessment unit; The inflammatory response analysis unit detects the gene expression patterns of IL-1β and TNF-α. When the IL-1β gene expression level increases to more than 4 times the basal level and the TNF-α gene expression level increases to more than 3 times the basal level, it is determined to be an acute inflammatory response state. When the IL-1β gene expression level increases to 2-4 times the basal level and the TNF-α gene expression level increases to 1.5-3 times the basal level, it is determined to be a chronic inflammatory response state. The immune function assessment unit conducts a comprehensive assessment of the activation level and functional status of the immune system based on the expression combination characteristics of inflammatory factor genes.

5. The aquaculture organism health status monitoring system based on multidimensional feature association according to claim 4 is characterized in that: The metabolic gene monitoring module includes a metabolic activity assessment unit and an energy balance analysis unit; The metabolic activity assessment unit monitors the expression changes of GLUT and FAS genes. When the GLUT gene expression level drops to less than 0.5 times the basal level and the FAS gene expression level drops to less than 0.3 times the basal level, it is judged as a severe metabolic inhibition state. When the GLUT gene expression level drops to 0.5-0.8 times the basal level and the FAS gene expression level drops to 0.3-0.7 times the basal level, it is judged as a moderate metabolic inhibition state. The energy balance analysis unit combines the expression characteristics of metabolism-related genes to dynamically monitor the body's energy metabolism level and nutritional status.

6. The aquaculture organism health status monitoring system based on multidimensional feature association according to claim 1 is characterized in that: The contents of the standardization module, feature screening module, deep learning module, and evaluation output module are as follows: The standardization module is used to perform Min-Max standardization and time-series alignment on the multi-dimensional data input from the visual feature acquisition layer and the molecular biology verification layer, achieving unified expression of heterogeneous data; The feature screening module is used to optimize and reduce the dimension of standardized features based on data correlation and information entropy theory, and dynamically sample time series features through a sliding window; The deep learning module uses an LSTM network structure to perform temporal modeling on the filtered features and implements correlation learning and pattern mining between visual features and molecular markers through a multi-layer attention mechanism; The evaluation output module is used to generate a health status score of 0-100 points and grade the scores based on preset thresholds to achieve visual early warning of the health status of farmed organisms.

7. The aquaculture organism health status monitoring system based on multidimensional feature association according to claim 1 is characterized in that: Also includes fault tolerance mechanisms: It includes data acquisition exception processing unit, equipment fault detection unit and system recovery unit; When the data collection exception processing unit detects that the video stream data collection is interrupted, it automatically switches to the backup camera to continue data collection, and records the time when the exception occurred and the duration; The equipment fault detection unit monitors the operating status and data transmission quality of each functional module. When a device fault or data transmission anomaly is detected, it automatically switches the relevant module to safe mode and issues an alarm message. The system recovery unit is responsible for data recovery and system parameter recalibration after troubleshooting, retrospectively evaluating the health status during the interruption by calling historical data, and performing an additional molecular biology verification within 72 hours after resuming normal operation.

8. The aquaculture organism health status monitoring system based on multidimensional feature association according to claim 7, characterized in that: It also includes an environmental adaptability adjustment mechanism, which automatically adjusts the camera's exposure parameters according to the light intensity and activates the infrared fill light mode when the light is lower than 100 lux; adjusts the threshold of the image preprocessing algorithm based on water quality parameters, and adds an image enhancement processing step when the water turbidity exceeds 50NTU; automatically corrects the evaluation criteria of behavioral characteristics according to changes in water temperature, and adjusts the baseline values of swimming speed and gill cover movement frequency accordingly when the water temperature is lower than 15℃ or higher than 30℃; automatically updates the calculation parameters of group behavioral characteristics when the breeding density changes, and triggers a molecular biological verification when the density changes by more than 30%.

9. The aquaculture organism health status monitoring system based on multidimensional feature association according to claim 6, characterized in that: The visual feature acquisition layer monitors fish behavior characteristics, including the following corresponding relationships: Swimming speed characteristics: By monitoring the displacement changes of fish per unit time, when the swimming speed is less than 50% of the normal value or the swimming trajectory shows irregular jitter, it indicates that the fish may be in a state of stress or abnormal energy metabolism. When the group swimming speed suddenly increases by more than 200% of the normal value, it indicates that the water quality may be deteriorating or there may be external interference. Clustering characteristics are determined by calculating the spatial distance distribution between individuals. When the nearest neighbor distance of more than 90% of the individuals is less than 30% of the normal value, it indicates that the fish are in a state of stress or fright. When the degree of individual dispersion increases significantly and the nearest neighbor distance is greater than 300% of the normal value, it indicates that there may be infectious diseases or insufficient dissolved oxygen in the water. The gill cover movement characteristics are measured by measuring the frequency of gill cover opening and closing. When the gill cover movement frequency exceeds 150% of the normal value and this state lasts for more than 2 hours, it indicates that the fish is in a state of difficulty breathing or insufficient dissolved oxygen; when the gill cover movement frequency of more than 50% of individuals is lower than 60% of the normal value, it indicates a decrease in metabolic activity or environmental stress.

Images