Fish disease intelligent identification method and system
By integrating multi-source data and using an intelligent identification system, the problems of data isolation and adaptation difficulties in traditional fishery disease prediction have been solved, enabling accurate disease prediction and low-cost dynamic adaptation to new environments and new fish species.
Patent Information
- Application Number
- CN202510958896.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-11
- Publication Date
- 2025-10-31
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Traditional methods for predicting fishery diseases suffer from problems such as isolated data dimensions, fragmented spatiotemporal relationships, distorted dynamic patterns, weak generalization ability, and high adaptation costs, especially poor adaptability across aquaculture farms and in new environments.
An intelligent identification system employing multi-source data acquisition, distributed storage, federated learning, and online learning, utilizes multi-task learning networks and blockchain technology to construct a global sub-model, achieving data fusion and adaptive optimization for disease prediction and water quality assessment.
It achieves more comprehensive causal analysis, more accurate dynamic prediction, and lower adaptation costs, adapting to new fish species and new environments, and provides a systematic solution for intelligent fishery disease prevention and control.
Smart Images
Figure CN120876140A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of fish disease identification technology, and in particular to a smart fish disease identification method and system. Background Technology
[0002] Predicting fishery diseases is a core aspect of ensuring aquaculture profitability, but traditional methods suffer from drawbacks such as isolated data dimensions, one-sided analysis of causes, fragmented spatiotemporal relationships and distorted dynamic patterns, as well as weak generalization ability and high adaptation costs. Because most existing models rely on a single data type, such as only water quality sensor data or only fish growth data, they ignore the synergistic effects of fish characteristics. For example, when the regional water temperature rises suddenly, the disease risk of cold-water fish is fundamentally different from that of warm-water fish. Furthermore, farms in the same watershed share similar environmental characteristics, but traditional models do not model regional commonalities, resulting in large prediction biases across farms. When new fish species or new farming environments are introduced, the models need to be retrained. The lack of a hierarchical design of "general features + specific adaptations" leads to poor adaptability to environmental changes and a lack of automatic update mechanisms. To address the aforementioned technical shortcomings, a solution is proposed. Summary of the Invention
[0003] The purpose of this invention is to solve the shortcomings of traditional methods, such as isolated data dimensions, one-sided causal analysis, fragmented spatiotemporal relationships and distorted dynamic patterns, as well as weak generalization ability and high adaptation costs.
[0004] To achieve the above objectives, the present invention adopts the following technical solution: A method for intelligent identification of fish diseases includes the following steps: Step 1: The multi-source data acquisition module collects aquaculture information, including fish disease image data, water quality sensor data, fish growth data, and diagnostic medication data. Step two, the distributed data storage module performs standardized processing and distributed storage of aquaculture information: Specifically, the aquaculture information is standardized through a preprocessing model, and the preprocessed fish disease image data and water quality sensor data are stored in a distributed file system; the preprocessed fish growth data and diagnostic drug data are stored in a distributed database, and the diagnostic drug data is stored using blockchain. Step 3: The central server establishes an information processing model. The central server includes a federated learning module and an online learning module. The federated learning module establishes a global sub-model of the aquaculture farm to analyze the impact of regional environmental changes on fish diseases. The online learning module performs self-growing optimization and updates on the global sub-model of the aquaculture farm. Step 4: The disease prediction and optimization module trains reasoning on fish disease conditions: By constructing a general model and a multi-task learning network, fish diseases are pre-trained and optimized respectively, so as to adapt to new fish species and new aquaculture environment, optimize aquaculture density and water quality control, and output disease prediction results, water quality assessment results and aquaculture optimization plan. Step 5: Use the interactive early warning module to view the results of the inference training and perform early warning processing: Access the system through the user terminal to view the disease prediction results and water quality assessment results, and receive aquaculture optimization plans for early warning processing.
[0005] Furthermore, real-time collection of aquaculture information is achieved through cameras, sensors, and smart devices deployed at aquaculture farms; Aquaculture information includes fish disease image data, water quality sensor data, fish growth data, and diagnostic medication data; Fish disease image data includes fish species, disease type, and lesion location; Water quality sensor data includes pH value, dissolved oxygen content, water temperature, conductivity, turbidity, and compound content; Fish growth data include body length, weight, growth cycle, and food intake; Diagnostic medication data includes the type of fish disease, the name and dosage of the medication, and the frequency and timing of medication administration.
[0006] Furthermore, the preprocessing model standardizes the aquaculture information, specifically through the following steps: Preprocessing fish disease image data: underwater shooting noise is removed by adaptive median filtering algorithm and pixel size is unified. Then, pixel values are standardized by Z-score, and texture and morphological features of the image are extracted at the same time. Preprocessing water quality sensor data: The sliding window algorithm is used to detect and remove outliers caused by sensor drift, then the missing data is filled in by interpolation and the parameter indicators are normalized, and then standard time series data is generated by aligning the data with timestamps. Preprocessing fish growth data: By unifying the units of measurement and establishing a standardized comparison table for growth indicators of different fish species, extreme values of fish growth data are then filtered out. Preprocessing diagnostic medication data: Standardize drug names, standardize drug dosage units, and convert them into structured labels.
[0007] Furthermore, the diagnostic medication data is stored using blockchain technology, specifically through the following steps: Structured data is generated from standardized diagnostic medication data, then encrypted into hash values using the SHA-256 algorithm, and GPS coordinates are added as unique identifiers for aquaculture ponds, thus performing data preprocessing before uploading to the blockchain. The central server serves as the consensus node of the consortium blockchain, and the local nodes of each farm serve as the accounting nodes. A new block is generated periodically, and the block contains the hash value of the previous block, the hash list of the current batch of data, and the timestamp, thus building a blockchain architecture. By assigning a unique digital fingerprint to each piece of data, including the block height and transaction ID, users can query the complete flow path by entering the fingerprint on their terminal, thereby establishing a traceability mechanism.
[0008] Furthermore, the specific process of establishing a global sub-model for the farm is as follows: The input is standardized aquaculture information, and the probability of disease risk is analyzed. The specific process is as follows: Fish disease image data features are labeled as Fa, and the lesion risk index LRI is obtained: fish species sensitivity weight Wsp is assigned according to fish species type, and organ importance weight Wps is assigned according to lesion location. The lesion proportion Slf is calculated by the ratio of lesion area to fish body surface area. The environmental correction coefficient Cenv is obtained by light intensity parameter and depth parameter. Then, the lesion risk index LRI is obtained by comprehensive analysis. Label the water quality sensor data features as Fb and obtain the water quality stress index WSI: label the total number of all parameters of the water quality sensor data as m, and label any one parameter of the water quality sensor data as k; Set a sliding window L, calculate the mean Xk of parameter index k in the sliding window L, set and mark the safe threshold Xks and dangerous threshold Xkm of parameter index k, and assign the correlation weight between parameter index and disease to Wk, and then comprehensively obtain the water quality stress index WSI. Fish growth data characteristics are labeled as Fc, and the growth resistance index GRI is obtained: The body length growth rate H1 is obtained by comparing the daily average increase in body length with the initial body length; the standard deviation of N0 individual body length growth rate H1s is used to obtain the body length growth fluctuation coefficient h1; the body weight growth rate H2 is obtained by comparing the daily average increase in body weight with the initial body weight; the standard deviation of N0 body weight growth rate H2s is used to obtain the body weight growth fluctuation coefficient h2; the feeding efficiency J1 is obtained by comparing the current food intake with body weight; and the condition factor J2 is calculated by comparing the current body weight with body length; thus, the growth resistance index GRI is obtained comprehensively. Label the diagnostic medication data characteristics as Fd, and obtain the medication risk index DRI: label the historical medication frequency as N1, the dosage of the i-th medication as Di, and the duration of the i-th medication as Ti, and set and label the drug efficacy decay coefficient as... Mark the time since medication use as This allows for the comprehensive acquisition of the Drug Risk Index (DRI). Then, correlation analysis was performed using the Pearson correlation coefficient to predetermine the weighting factor coefficients of the Lesion Risk Index (LRI), Water Quality Stress Index (WSI), Growth Resistance Index (GRI), and Drug Use Risk Index (DRI), and these coefficients were labeled accordingly. , , , Then, a disease risk probability score P is generated by combining the results. Set an assessment interval for the disease risk probability score P, and assess the disease risk level by comparing intervals.
[0009] Furthermore, the specific process of constructing a general model for pre-training against fish diseases is as follows: The model extracts general visual features from fish disease images using a pre-trained ResNet-50 model, and then connects them in parallel to a multi-source environmental feature encoder. The model is pre-trained using historical datasets across fish species and environments. The training objective is to minimize the disease classification loss. By calculating the cross-entropy loss, the model obtains and outputs disease prediction results, including disease type and risk level.
[0010] Furthermore, the specific process of constructing a multi-task learning network is as follows: By optimizing the training for different types of fish diseases, the MAE loss function of the water quality stress index (WSI) and the Huber loss function with respect to stocking density were calculated. Then, by dynamically adjusting the weights of the loss function, the lesion risk index LRI is minimized, thereby optimizing the training strategy. Furthermore, by optimizing training strategies, water quality assessment results and aquaculture optimization plans can be obtained, thereby optimizing aquaculture density and water quality control.
[0011] A fish disease intelligent identification system includes a multi-source data acquisition module, a distributed data storage module, a central server, a disease prediction and optimization module, and an application interaction and early warning module. The central server includes a federated learning module and an online learning module, and the modules are interconnected. When the system is applied, it executes the aforementioned fish disease intelligent identification method. The multi-source data acquisition module is used to collect information on aquaculture. The distributed data storage module is used for standardized processing and distributed storage of aquaculture information; The central server is used to build an information processing model; among them, the federated learning module builds a global sub-model of the farm, and the online learning module performs self-growing optimization and updates on the global sub-model of the farm. The disease prediction and optimization module is used for reasoning and training on fish disease conditions. The application interaction warning module is used to view the results of inference training schemes and to issue warnings.
[0012] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: This invention integrates multi-dimensional data through a closed-loop design of multi-source data fusion, spatiotemporal modeling, generalization adaptation, and interpretable decision-making, making the causal analysis more comprehensive. It also designs a spatiotemporal modeling architecture to accurately capture dynamic patterns and generalize and self-updating hierarchical transfer learning, thereby significantly reducing adaptation costs. This invention integrates cross-type data, combines water quality time-series dynamics, regional environmental statics and dynamics, and deeply couples fish characteristics to construct a chain-like causal model of "environment → aquaculture pond → fish body," and conducts interactive feature mining to quantify the synergistic effects of cross-level factors through interactive features. This invention captures time-dependent effects by identifying local anomaly patterns in water quality data, reconstructs the evolutionary process of diseases from "gradual change to sudden change," and improves regional prediction accuracy by distinguishing between environmental commonalities and farm-specific characteristics in different watersheds through regional ID coding and attention fusion. This invention learns the common disease characteristics across fish species and regions, assesses the correlation between fish disease lesions and water quality abnormalities, thereby adapting to new fish species and new aquaculture environments, and establishing an incremental self-updating mechanism to quickly respond to environmental changes; This invention addresses the pain points of traditional fishery disease prediction, such as data isolation, distorted patterns, difficulty in adaptation, and blind decision-making, through a closed-loop design of multi-source data fusion → spatiotemporal modeling → generalization adaptation → interpretable decision-making → edge deployment. It achieves more comprehensive causal analysis, more accurate dynamic prediction, lower adaptation costs, and more direct decision support, providing a systematic solution for disease prevention and control in smart fisheries. Attached Figure Description
[0013] Figure 1 A schematic diagram of the steps in the method flow of the present invention is shown; Figure 2 A schematic diagram of the system modules of the present invention is shown. Detailed Implementation
[0014] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0015] Example 1:
[0016] like Figures 1-2As shown, a fish disease intelligent identification system includes a multi-source data acquisition module, a data distributed storage module, a central server, a disease prediction and optimization module, and an application interaction and early warning module. The central server includes a federated learning module and an online learning module, and the modules are connected to each other. The work steps are as follows: S1, the multi-source data acquisition module collects aquaculture information: aquaculture information includes fish disease image data, water quality sensor data, fish growth data, and diagnostic medication data; Real-time collection of aquaculture information is achieved through cameras, sensors, and smart devices deployed at aquaculture farms. Aquaculture information includes fish disease image data, water quality sensor data, fish growth data, and diagnostic medication data; Fish disease image data includes fish species, disease type, and lesion location; Among them, high-definition cameras are installed in the breeding ponds to take pictures at regular intervals, and drones are used to take aerial pictures of the areas where diseased fish gather. Then, close-up pictures of diseased fish are taken by sampling, so as to manually record the fish species, disease type, and lesion location. The light intensity and depth of the underwater location are recorded by intelligent sensors. Water quality sensor data includes pH value, dissolved oxygen content, water temperature, conductivity, turbidity, and compound content; The system utilizes multi-parameter integrated sensors for immersion data acquisition, including pH, conductivity, and ammonia nitrogen detection modules. These sensors are installed at different depths in the surface, middle, and bottom layers of the aquaculture pond and transmit data to local nodes via wireless communication. Fish growth data include body length, weight, growth cycle, and food intake; Records were kept through sampling and testing, using calipers and electronic scales to measure body length and weight; food intake was measured using an automatic feeder, and the growth cycle was recorded manually. Diagnostic medication data includes the type of fish disease, the name and dosage of the medication, and the frequency and timing of medication administration. Among them, diagnostic medication parameters are entered through the terminal of the aquaculture personnel and automatically converted to the same dimension.
[0017] S2, the distributed data storage module performs standardized processing and distributed storage of aquaculture information: Specifically, the preprocessing model standardizes aquaculture information, stores preprocessed fish disease image data and water quality sensor data in a distributed file system; and stores preprocessed fish growth data and diagnostic medication data in a distributed database, and uses blockchain to store diagnostic medication data to ensure that the data is tamper-proof and traceable. S2-1, The preprocessing model standardizes aquaculture information, specifically through the following steps: Preprocessing fish disease image data: underwater shooting noise is removed by adaptive median filtering algorithm and pixel size is unified. Then, pixel values are standardized by Z-score, and texture and morphological features of the image are extracted at the same time. Texture features include grayscale LBP feature maps, and morphological features include lesion region contour maps. The grayscale LBP feature maps are extracted using existing image processing algorithms, and the lesion region contour maps are labeled by professional veterinarians to form a sample database of lesion images. Preprocessing water quality sensor data: The sliding window algorithm is used to detect and remove outliers caused by sensor drift. Then, the missing data is filled in by interpolation and the parameter index is normalized to the [0,1] interval. Finally, standard time series data is generated by aligning the data with the timestamp. Preprocessing fish growth data: By using a unified unit of measurement and establishing a standardized comparison table for growth indicators of different fish species, the body length growth rate is obtained by the ratio of the average daily growth value to the initial body length, and the body condition is calculated by the ratio of body weight to body length. Then, extreme values of fish growth data, such as abnormal body weight data caused by sudden death, are filtered out. Preprocessing diagnostic medication data: Standardize drug names through the veterinary pharmacopoeia, standardize drug dosage units, and convert text information such as medication time, fish disease type, and medication effect into structured tags, including medication time, drug information, dosage, operator ID, fish disease type, and post-medication effect; By using a preprocessing model, fishery aquaculture information is standardized in a targeted manner to achieve data format unification, noise filtering, and spatiotemporal alignment of features; S2-2, using blockchain to store diagnostic medication data, the specific operation is as follows: Structured data is generated from standardized diagnostic medication data, then encrypted into hash values using the SHA-256 algorithm, and GPS coordinates are added as unique identifiers for aquaculture ponds, thus performing data preprocessing before uploading to the blockchain. The central server serves as the consensus node of the consortium blockchain, and the local nodes of each farm serve as the accounting nodes. The PBFT practical Byzantine fault-tolerant consensus mechanism is adopted to generate a new block at regular intervals. The block contains the hash value of the previous block, the hash list of the current batch of data, and the timestamp, thereby building the blockchain architecture. By using chained hash association, each block hash depends on the previous block, ensuring that data modification requires the reconstruction of all subsequent blocks. Furthermore, smart contracts are used to pre-set data writing rules, such as allowing only authorized personnel to upload data and preventing data from being deleted after upload, thus ensuring that the data cannot be tampered with. By assigning a unique digital fingerprint to each piece of data, including block height and transaction ID, users can query the complete flow path by entering the fingerprint on the terminal, and trace historical data by dimensions such as time, fish species, and drug type, thereby establishing a traceability mechanism and realizing the full life cycle evidence storage of diagnostic medication data.
[0018] S3, The central server establishes an information processing model: The central server includes a federated learning module and an online learning module. The federated learning module establishes a global sub-model of the aquaculture farm and builds a communication architecture between the central server and local nodes of each aquaculture farm, thereby analyzing the impact of regional environmental changes on fish diseases; the online learning module performs self-growing optimization and updates on the global sub-model of the aquaculture farm. S3-1, The specific process of establishing the global sub-model of the farm is as follows: The input is standardized aquaculture information, and the probability of disease risk is analyzed. The specific process is as follows: S3-101, the fish disease image data features are labeled as Fa, and the lesion risk index LRI is obtained; Assign a corresponding sensitivity weight Wsp to the fish species. For example, grass carp are sensitive to gill rot, so the preset sensitivity weight is high; bass are tolerant to gill rot, so the preset sensitivity weight is low. Assign corresponding organ importance weights (Wps) to the location of the lesion. For example, the importance of the cheek is higher than that of the body surface, which is higher than that of the tail. The preset values of the corresponding organ importance weights are from high to low. The proportion of lesions (Slf) is calculated by the ratio of the lesion area to the fish's body surface area. The environmental correction coefficient Cenv is obtained using light intensity and depth parameters. ;
[0019] Among them, the light intensity parameter includes the actual light intensity and the standard light intensity, and the depth parameter includes the actual shooting depth and the standard depth; This represents the actual light intensity. Standard light intensity; To capture the actual depth, For example, if 500 lux is chosen as the standard light intensity, then 1m is chosen as the standard depth. and The influence weights of light intensity and depth parameters are respectively obtained through manual experience presets. and ; Furthermore, the lesion risk index (LRI) is obtained through comprehensive analysis: ;
[0020] When the fish species sensitivity weight Wsp, the lesion organ importance weight Wps, the lesion proportion Slf, and the environmental correction coefficient Cenv are higher, the lesion risk index LRI is higher. S3-102, mark the water quality sensor data features as Fb, and obtain the water quality stress index WSI; The total number of all parameters in the water quality sensor data is labeled as m, and any one parameter in the water quality sensor data is labeled as k. Set a sliding window L and calculate the mean Xk of parameter index k within the sliding window L; Set and label the safety threshold Xks and danger threshold Xkm of parameter index k, and assign the correlation weight between parameter index and disease to Wk; Then, the Water Stress Index (WSI) is obtained by comprehensive analysis: ; The higher the correlation weight Wk between the parameter index and the disease, the higher the difference between the mean Xk of the parameter index k and the safe threshold Xks, and the lower the difference between the danger threshold Xkm and the safe threshold Xks, the higher the water quality stress index WSI. S3-103, fish growth data features are labeled as Fc, and the growth resistance index GRI is obtained; The body length growth rate H1 is obtained by comparing the average daily increase in body length with the initial body length. The standard deviation of the body length growth rate H1 for N0 individuals is calculated to obtain the body length growth fluctuation coefficient h1. The weight growth rate H2 is obtained by comparing the average daily weight gain with the initial weight. The standard deviation of N0 weight growth rate H2s is then calculated to obtain the weight growth fluctuation coefficient h2. The feeding efficiency J1 is obtained by the ratio of food intake to body weight at the current time point; The body fat percentage J2 is calculated by the ratio of body weight to body length at the current time point. Furthermore, the growth resistance index (GRI) is obtained through comprehensive analysis. ;
[0021] When the growth rate of body length H1 is higher and the coefficient of body length growth fluctuation h1 is lower, the growth rate of body weight H2 is higher and the coefficient of body weight growth fluctuation h2 is lower, and the feeding efficiency J1 and fatness J2 are higher, the growth resistance index GRI is higher. S3-104, label the characteristics of diagnostic medication data as Fd, and obtain the medication risk index DRI; Label the historical number of medication administrations as N1, the dosage of the i-th administration as Di, and the duration of the i-th administration as Ti. Set and label the drug efficacy decay coefficient as... Mark the time since medication use as ; Then, the Drug Risk Index (DRI) is obtained by comprehensively analyzing the medication risks. ; When the ratio of drug dose Di to drug duration Ti is higher, and the time since drug administration is longer... The lower the value, the higher the drug resistance and the greater the decline in drug efficacy, and the higher the drug risk index (DRI). S3-2, then build a communication architecture between the central server and local nodes in each aquaculture farm to analyze the impact of regional environmental changes on fish diseases; The central server establishes communication with the local nodes of the farm and the data transmission is encrypted. The local nodes complete the node identity authentication through digital certificates and group the farms by geographical region division. The central server aggregates the model parameters of the nodes in the same region and generates regional sub-models. The regional sub-model uses Pearson correlation coefficient analysis to predetermine the weighting factor coefficients of the Lesion Risk Index (LRI), Water Quality Stress Index (WSI), Growth Resistance Index (GRI), and Drug Use Risk Index (DRI), and labels them sequentially as follows: , , , ; Among them, the preset weight factor coefficients are all greater than 0, and the higher the Pearson coefficient, the higher the preset value of the weight factor coefficient. Then, a disease risk probability score P is generated by combining the results: ; When the Lesion Risk Index (LRI), Water Stress Index (WSI), and Drug Use Risk Index (DRI) are higher, and the Growth Resistance Index (GRI) is lower, the disease risk probability score (P) is higher. Set the assessment interval for the disease risk probability score P, and assess the disease risk level by comparing intervals; S3-3, the specific process of self-growing optimization and updating the global sub-model of the farm is as follows: S3-301, by re-collecting aquaculture information, the Pearson correlation coefficient is recalculated: Regularly collect aquaculture information data when there are significant changes in factors such as the aquaculture farm environment, recalculate the Pearson correlation coefficient, and analyze the correlation between each index and the occurrence of diseases; S3-302 resets the weight factor coefficients of the disease risk index LRI, water quality stress index WSI, growth resistance index GRI, and medication risk index DRI to assess the degree of influence of the index on disease risk, thereby performing self-growing optimization and updating of the global sub-model of the aquaculture farm. For example, if it is found that the correlation between recent water quality changes and disease occurrence has increased, that is, the Pearson correlation coefficient between the Water Quality Stress Index (WSI) and disease risk has increased, then the weight factor coefficient of the WSI should be increased accordingly.
[0022] S4, the disease prediction and optimization module performs reasoning training on fish disease conditions: by constructing a general model and a multi-task learning network, fish diseases are pre-trained and optimized respectively, so as to adapt to new fish species and new aquaculture environments, optimize aquaculture density and water quality control, and output disease prediction results, water quality assessment results and aquaculture optimization schemes. S4-1, The specific process of constructing a general model for pre-training on fish diseases is as follows: S4-101 extracts general visual features from fish disease images using a pre-trained ResNet-50 model, and connects to a multi-source environmental feature encoder in parallel to process structured data such as water temperature, dissolved oxygen, and aquaculture patterns. It also fuses image and environmental features through an attention mechanism. S4-102 is then pre-trained using historical datasets across fish species and environments. The training objective is to minimize the disease classification loss. The cross-entropy loss is calculated to evaluate the ability of the learning-enhanced model to distinguish features of similar fish species. S4-103, then obtain disease prediction results, so as to adapt to new fish species and new aquaculture environment; For new fish species: fine-tune the task adaptation layer using n0 labeled disease samples, and transfer the general features to the new fish species through transfer learning; For new aquaculture environments: By adding new environmental feature dimensions, such as salinity and alkalinity, and corresponding feature weight parameters, rapid iteration can be performed to improve prediction accuracy; S4-104 outputs disease prediction results, including disease type and risk level; S4-2, The specific process of constructing a multi-task learning network is as follows: S4-201 calculates the MAE loss function of the water quality stress index (WSI) and the Huber loss function with respect to stocking density by optimizing training for different types of fish diseases. S4-202, and then by dynamically adjusting the weights of the loss function, the lesion risk index LRI is minimized, thereby optimizing the training strategy; S4-203, and then obtains water quality assessment results and aquaculture optimization schemes through optimized training strategies, thereby optimizing aquaculture density and water quality control, and storing them in the data distributed storage module; The water quality assessment results include the compliance rate of specific indicators and the analysis of potential pollution sources; the aquaculture optimization plan is detailed, including water quality control measures and recommendations for adjusting stocking density, and is dynamically updated according to the fish growth cycle; Water quality control measures include adjusting the frequency and duration of aerator operation and the frequency of water changes; recommendations for adjusting stocking density include the number of stocked animals and the cycle of dividing the ponds.
[0023] S5, the application interactive early warning module can view the results of the inference training scheme and perform early warning processing: users can access the system through the user terminal to view the disease prediction results and water quality assessment results, and receive aquaculture optimization schemes for early warning processing.
[0024] In summary, this invention integrates multi-dimensional data through a closed-loop design of multi-source data fusion, spatiotemporal modeling, generalization adaptation, and interpretable decision-making, making the causal analysis more comprehensive. It also designs a spatiotemporal modeling architecture to accurately capture dynamic patterns and achieves generalization and self-updating hierarchical transfer learning, thereby significantly reducing adaptation costs. This invention integrates cross-type data, combines water quality time-series dynamics, regional environmental statics and dynamics, and deeply couples fish characteristics to construct a chain-like causal model of "environment → aquaculture pond → fish body," and conducts interactive feature mining to quantify the synergistic effects of cross-level factors through interactive features. This invention captures time-dependent effects by identifying local anomaly patterns in water quality data, reconstructs the evolutionary process of diseases from "gradual change to sudden change," and improves regional prediction accuracy by distinguishing between environmental commonalities and farm-specific characteristics in different watersheds through regional ID coding and attention fusion. This invention learns the common disease characteristics across fish species and regions, assesses the correlation between fish disease lesions and water quality abnormalities, thereby adapting to new fish species and new aquaculture environments, and establishing an incremental self-updating mechanism to quickly respond to environmental changes; This invention addresses the pain points of traditional fishery disease prediction, such as data isolation, distorted patterns, difficulty in adaptation, and blind decision-making, through a closed-loop design that integrates multi-source data fusion, spatiotemporal modeling, generalization adaptation, interpretable decision-making, and edge deployment. It achieves more comprehensive causal analysis, more accurate dynamic prediction, lower adaptation costs, and more direct decision support, providing a systematic solution for disease prevention and control in smart fisheries.
[0025] 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 in 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.
[0026] 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; that is, 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 according to actual needs.
[0027] The size of the interval and threshold is set to facilitate comparison. The size of the threshold depends on the amount of sample data and the number of bases set by those skilled in the art for each set of sample data; as long as it does not affect the ratio between the parameter and the quantized value.
[0028] 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.
[0029] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.
Claims
1. A method for intelligent identification of fish diseases, characterized in that: Includes the following steps: Step 1: The multi-source data acquisition module collects aquaculture information, including fish disease image data, water quality sensor data, fish growth data, and diagnostic medication data. Step two, the distributed data storage module performs standardized processing and distributed storage of aquaculture information: Specifically, the aquaculture information is standardized through a preprocessing model, and the preprocessed fish disease image data and water quality sensor data are stored in a distributed file system; the preprocessed fish growth data and diagnostic drug data are stored in a distributed database, and the diagnostic drug data is stored using blockchain. Step 3: The central server establishes an information processing model. The central server includes a federated learning module and an online learning module. The federated learning module establishes a global sub-model of the aquaculture farm to analyze the impact of regional environmental changes on fish diseases. The online learning module performs self-growing optimization and updates on the global sub-model of the aquaculture farm. Step 4: The disease prediction and optimization module trains reasoning on fish disease conditions: By constructing a general model and a multi-task learning network, fish diseases are pre-trained and optimized respectively, so as to adapt to new fish species and new aquaculture environment, optimize aquaculture density and water quality control, and output disease prediction results, water quality assessment results and aquaculture optimization plan. Step 5: Use the interactive early warning module to view the results of the inference training and perform early warning processing: Access the system through the user terminal to view the disease prediction results and water quality assessment results, and receive aquaculture optimization plans for early warning processing.
2. The intelligent fish disease identification method according to claim 1, characterized in that: Real-time collection of aquaculture information is achieved through cameras, sensors, and smart devices deployed at aquaculture farms. Aquaculture information includes fish disease image data, water quality sensor data, fish growth data, and diagnostic medication data; Fish disease image data includes fish species, disease type, and lesion location; Water quality sensor data includes pH value, dissolved oxygen content, water temperature, conductivity, turbidity, and compound content; Fish growth data include body length, weight, growth cycle, and food intake; Diagnostic medication data includes the type of fish disease, the name and dosage of the medication, and the frequency and timing of medication administration.
3. The intelligent fish disease identification method according to claim 2, characterized in that: The preprocessing model standardizes aquaculture information, specifically by: Preprocessing fish disease image data: underwater shooting noise is removed by adaptive median filtering algorithm and pixel size is unified. Then, pixel values are standardized by Z-score, and texture and morphological features of the image are extracted at the same time. Preprocessing water quality sensor data: The sliding window algorithm is used to detect and remove outliers caused by sensor drift, then the missing data is filled in by interpolation and the parameter indicators are normalized, and then standard time series data is generated by aligning the data with timestamps. Preprocessing fish growth data: By unifying the units of measurement and establishing a standardized comparison table for growth indicators of different fish species, extreme values of fish growth data are then filtered out. Preprocessing diagnostic medication data: Standardize drug names, standardize drug dosage units, and convert them into structured labels.
4. The intelligent fish disease identification method according to claim 3, characterized in that: The specific steps for storing diagnostic medication data using blockchain are as follows: Structured data is generated from standardized diagnostic medication data, then encrypted into hash values using the SHA-256 algorithm, and GPS coordinates are added as unique identifiers for aquaculture ponds, thus performing data preprocessing before uploading to the blockchain. The central server serves as the consensus node of the consortium blockchain, and the local nodes of each farm serve as the accounting nodes. A new block is generated periodically, and the block contains the hash value of the previous block, the hash list of the current batch of data, and the timestamp, thus building a blockchain architecture. By assigning a unique digital fingerprint to each piece of data, including the block height and transaction ID, users can query the complete flow path by entering the fingerprint on their terminal, thereby establishing a traceability mechanism.
5. The intelligent fish disease identification method according to claim 4, characterized in that: The specific process of establishing a global sub-model of the farm is as follows: The input is standardized aquaculture information, and the probability of disease risk is analyzed. The specific process is as follows: Fish disease image data features are labeled as Fa, and the lesion risk index LRI is obtained: fish species sensitivity weight Wsp is assigned according to fish species type, and organ importance weight Wps is assigned according to lesion location. The lesion proportion Slf is calculated by the ratio of lesion area to fish body surface area. The environmental correction coefficient Cenv is obtained by light intensity parameter and depth parameter. Then, the lesion risk index LRI is obtained by comprehensive analysis. Label the water quality sensor data features as Fb and obtain the water quality stress index WSI: label the total number of all parameters of the water quality sensor data as m, and label any one parameter of the water quality sensor data as k; Set a sliding window L, calculate the mean Xk of parameter index k in the sliding window L, set and mark the safe threshold Xks and dangerous threshold Xkm of parameter index k, and assign the correlation weight between parameter index and disease to Wk, and then comprehensively obtain the water quality stress index WSI. Fish growth data characteristics are labeled as Fc, and the growth resistance index GRI is obtained: The body length growth rate H1 is obtained by comparing the daily average increase in body length with the initial body length; the standard deviation of N0 individual body length growth rate H1s is used to obtain the body length growth fluctuation coefficient h1; the body weight growth rate H2 is obtained by comparing the daily average increase in body weight with the initial body weight; the standard deviation of N0 body weight growth rate H2s is used to obtain the body weight growth fluctuation coefficient h2; the feeding efficiency J1 is obtained by comparing the current food intake with body weight; and the condition factor J2 is calculated by comparing the current body weight with body length; thus, the growth resistance index GRI is obtained comprehensively. Label the diagnostic medication data characteristics as Fd, and obtain the medication risk index DRI: label the historical medication frequency as N1, the dosage of the i-th medication as Di, and the duration of the i-th medication as Ti, and set and label the drug efficacy decay coefficient as... Mark the time since medication use as This allows for the comprehensive acquisition of the Drug Risk Index (DRI). Then, correlation analysis was performed using the Pearson correlation coefficient to predetermine the weighting factor coefficients of the Lesion Risk Index (LRI), Water Quality Stress Index (WSI), Growth Resistance Index (GRI), and Drug Use Risk Index (DRI), and these coefficients were labeled accordingly. , , , Then, a disease risk probability score P is generated by combining the results. Set an assessment interval for the disease risk probability score P, and assess the disease risk level by comparing intervals.
6. The intelligent fish disease identification method according to claim 5, characterized in that: The specific process of constructing a general model for pre-training on fish diseases is as follows: The model extracts general visual features from fish disease images using a pre-trained ResNet-50 model, and then connects them in parallel to a multi-source environmental feature encoder. The model is pre-trained using historical datasets across fish species and environments. The training objective is to minimize the disease classification loss. By calculating the cross-entropy loss, the model obtains and outputs disease prediction results, including disease type and risk level.
7. The intelligent fish disease identification method according to claim 6, characterized in that: The specific process of constructing a multi-task learning network is as follows: By optimizing the training for different types of fish diseases, the MAE loss function of the water quality stress index (WSI) and the Huber loss function with respect to stocking density were calculated. Then, by dynamically adjusting the weights of the loss function, the lesion risk index LRI is minimized, thereby optimizing the training strategy. Furthermore, by optimizing training strategies, water quality assessment results and aquaculture optimization plans can be obtained, thereby optimizing aquaculture density and water quality control.
8. A fish disease intelligent identification system, characterized in that: The system includes a multi-source data acquisition module, a distributed data storage module, a central server, a disease prediction and optimization module, and an application interaction and early warning module. The central server includes a federated learning module and an online learning module, and the modules are interconnected. When the system is applied, it executes the fish disease intelligent identification method described in any one of claims 1-7. The multi-source data acquisition module is used to collect information on aquaculture. The distributed data storage module is used for standardized processing and distributed storage of aquaculture information; The central server is used to build an information processing model; among them, the federated learning module builds a global sub-model of the farm, and the online learning module performs self-growing optimization and updates on the global sub-model of the farm. The disease prediction and optimization module is used for reasoning and training on fish disease conditions. The application interaction warning module is used to view the results of inference training schemes and to issue warnings.