An intelligent inspection system and method for fish diseases in aquaculture

Abstract

This invention provides an intelligent inspection system and method for aquaculture diseases, including a fish disease information collection device and a fish disease identification intelligent agent subsystem connected to it via the Internet. The fish disease information collection device includes an intelligent inspection device located within the aquaculture facility and a data transmission device and industrial control computer located in a control cabinet. The fish disease identification intelligent agent subsystem includes a fish disease identification model, a water quality-linked diagnostic module, and an incremental learning framework. This invention has the following advantages: the fish disease identification intelligent agent in this invention is based on an improved deep learning algorithm and combines computer vision and multimodal data fusion technology, enabling fish species identification, physical signs and health assessment, disease diagnosis, etiology analysis, and prevention and control strategy generation. This solves the problems of traditional aquaculture disease diagnosis relying on human experience, low accuracy, and delayed response, and enables intelligent diagnosis and control of fish diseases.

Description

Technical Field

[0001] This invention relates to the field of aquaculture disease monitoring technology, and in particular to an intelligent inspection system and method for aquaculture fish diseases. Background Technology

[0002] In aquaculture, timely diagnosis and control of fish diseases are crucial to the success of the farming operation. Traditional disease diagnosis in aquaculture relies heavily on manual experience, with farmers visually observing fish swimming postures and body surface characteristics to determine disease status. This method has several drawbacks: first, it has low accuracy, easily influenced by the experience level and subjective judgment of the personnel; second, it is slow to respond, failing to provide early warnings and often only discovering diseases after a large-scale outbreak; and third, manual inspections cannot cover the complex underwater environment, making it impossible to conduct precise monitoring of fish vital signs or integrate with water quality parameters for comprehensive diagnosis. As aquaculture develops towards large-scale and intelligent operations, traditional manual diagnostic methods can no longer meet industry needs, necessitating an integrated inspection system capable of accurate disease identification, early warning, and intelligent control. Summary of the Invention

[0003] In view of the shortcomings of the prior art described above, the purpose of this invention is to provide an intelligent inspection system and method for aquaculture diseases, which solves the problems of low accuracy in identifying fish diseases, inability to achieve early warning of diseases, and inability to perform refined monitoring of fish vital signs in the prior art.

[0004] To achieve the above and other related objectives, the present invention provides the following technical solution:

[0005] A smart inspection system for aquaculture diseases includes a fish disease information collection device and a fish disease identification intelligent subsystem connected to it via the Internet. The fish disease information collection device includes a smart inspection device located within the aquaculture facility and a data transmission device and industrial control computer located in a control cabinet. The industrial control computer is connected to the smart inspection device and the fish disease identification intelligent subsystem via the data transmission device. The smart inspection device includes a housing, a walking mechanism mounted on the housing and slidably connected to a closed track on the aquaculture facility, a lifting mechanism mounted on the side of the housing away from the walking mechanism, and a detection platform mounted on the lifting mechanism via a first support frame.

[0006] In one embodiment of the present invention, the walking mechanism includes a walking wheel slidably connected to a closed track disposed on a breeding facility and a drive assembly for driving the walking wheel to rotate. The drive assembly includes a drive motor and a reducer located within the housing. The output shaft of the drive motor is connected to the walking wheel via a transmission shaft.

[0007] In one embodiment of the present invention, the detection platform includes a second support frame connected to the lifting mechanism via a first support frame, and a camera assembly, a water quality sensor assembly, and a light source mounted on the second support frame. The camera assembly includes a first camera for acquiring images of fish at different water layers within the aquaculture facility and a second camera for acquiring water color images and spectral data of the water within the aquaculture facility.

[0008] A method for intelligent inspection of aquaculture fish diseases, based on the aforementioned intelligent inspection system for aquaculture fish diseases and applied to a fish disease identification intelligent agent subsystem, wherein the fish disease identification intelligent agent subsystem includes a fish disease identification model, a water quality linkage diagnosis module, and an incremental learning framework. The method includes the following steps: acquiring fish images at different water layers in the aquaculture facility captured by a first camera, as well as water quality parameters and aquaculture logs related to the aquaculture facility, and preprocessing the fish images at different water layers in the aquaculture facility;

[0009] Fish features are extracted from the preprocessed fish images, and multimodal features are obtained based on the fish features extracted from the fish images, water quality parameters, and aquaculture logs. The multimodal features are then input into a fish disease identification model, which is used to identify the species of fish and diagnose diseases in the aquaculture facility.

[0010] In one embodiment of the present invention, before acquiring fish images of different water layers in the aquaculture facility collected by the first camera, as well as water quality parameters and aquaculture logs of the aquaculture facility, and before preprocessing the fish images of different water layers in the aquaculture facility, the method further includes: acquiring water color images and spectral data of the aquaculture facility collected by the second camera, and preprocessing the water color images and spectral data of the aquaculture facility.

[0011] Features are extracted from the preprocessed water color image and spectral data, and the features are input into the water quality AI inversion model. The water quality parameters of the aquaculture facility are obtained through the water quality AI inversion model. The water quality parameters are compared with the preset fishery water quality standard threshold, and the water quality environment in the aquaculture facility is judged as abnormal and risk-classified according to the comparison results.

[0012] In one embodiment of the present invention, the step of extracting fish body features from the preprocessed fish body image and obtaining multimodal features based on the fish body features extracted from the fish body image, water quality parameters, and aquaculture logs includes: extracting fish body features from the preprocessed fish body image using lightweight depthwise separable convolution and an improved channel attention mechanism, extracting features from the water quality parameters and aquaculture logs, fusing the fish body features with the features from the water quality parameters and aquaculture logs to perform multimodal feature fusion, and obtaining multimodal features based on the fusion result.

[0013] In one embodiment of the present invention, obtaining the species identification and disease diagnosis of fish in the aquaculture facility through the fish disease identification model includes: the fish disease identification model uses an improved CIoU loss to complete fish localization, lesion detection and physical sign change identification, and combines cross-entropy loss to obtain the species identification and disease diagnosis of fish in the aquaculture facility.

[0014] In one embodiment of the present invention, after inputting the multimodal features into the fish disease identification model and obtaining the species identification and disease diagnosis of the fish in the aquaculture facility through the fish disease identification model, the method further includes: automatically matching the corresponding regulatory agents and treatment methods from the preset drug library, drug use rule library, and prevention and control scheme library based on the species identification and disease diagnosis of the fish in the aquaculture facility and the abnormal judgment and risk classification of the water quality environment in the aquaculture facility, and calculating the drug concentration and dosage of the corresponding regulatory agents.

[0015] In one embodiment of the present invention, the method further includes the following steps: collecting new fish disease sample data through an incremental learning framework, fine-tuning the network parameters in the fish disease identification model using the new fish disease sample data, and combining historical diagnostic knowledge to obtain an optimized fish disease identification model; verifying the accuracy of the optimized fish disease identification model, and iteratively adjusting it until the accuracy of the fish disease identification model meets the requirements, thereby improving the adaptability and robustness of the fish disease identification model.

[0016] As described above, the intelligent inspection system and method for farmed fish diseases of the present invention have the following beneficial effects:

[0017] 1. This invention can identify fish species: by determining the species of fish based on information such as body shape, color, fin shape, habits, and group structure; 2. This invention can assess the health of fish: based on physical characteristics such as body color, physical signs, body surface mucus, scale condition, gill condition, swimming posture, and eye condition, the health status of fish can be assessed; 3. This invention can diagnose fish diseases: by determining whether fish have potential parasites, bacterial or viral infections, water quality stress syndrome, or other types of diseases; 4. This invention can analyze the causes of fish diseases: by analyzing the causes of diseases from the dimensions of water quality, stocking density, nutrition, stress response, and pathogen introduction; 5. This invention can generate prevention and control strategies: by generating potential risk warnings and aquaculture optimization suggestions, providing targeted prevention and control solutions. Attached Figure Description

[0018] Figure 1 The diagram shows the overall structure of the intelligent fish disease inspection system according to the first embodiment of the present invention in three-dimensional form.

[0019] Figure 2The diagram shows the overall structure of the intelligent disease inspection system for farmed fish according to the first embodiment of the present invention.

[0020] Figure 3 The flowchart shown is the overall process flow of the intelligent inspection method for farmed fish diseases in the second embodiment of the present invention.

[0021] Figure 4 The flowchart shows the fish disease identification and prevention process of the intelligent inspection method for farmed fish diseases in the second embodiment of the present invention in practical application.

[0022] Figure 5 The diagram shows the overall flowchart of the intelligent inspection method for farmed fish diseases according to the second embodiment of the present invention in practical application.

[0023] Component designation explanation

[0024] 1. Housing; 2. Wheels; 3. Drive assembly; 4. Enclosed track; 5. UWB positioning tag; 6. Alarm light; 7. Lifting mechanism; 8. Second support frame; 9. First camera; 10. Second camera; 11. Water quality sensor assembly; 12. Light source. Detailed Implementation

[0025] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. It should be noted that, unless otherwise specified, the following embodiments and features described herein can be combined with each other.

[0026] The first embodiment of the present invention relates to an intelligent inspection system for farmed fish diseases, specifically as follows: Figure 1 and Figure 2 As shown, the invention includes a fish disease information collection device and a fish disease identification intelligent body subsystem. The fish disease information collection device and the fish disease identification intelligent body subsystem are connected through a network communication module. Furthermore, the invention adopts an integrated "cloud-edge-device" architecture and is configured with edge computing nodes, enabling it to operate independently in areas with poor network conditions.

[0027] The fish disease information collection device includes an intelligent inspection device located within the aquaculture facility and a data transmission device and industrial control computer located in the control cabinet. The industrial control computer is connected to the intelligent inspection device and the fish disease identification intelligent subsystem respectively through the data transmission device. In this embodiment, the industrial control computer is a small waterproof data processor. The industrial control computer uses a 12 / 24VDC power supply, has cyclic storage and cloud synchronization support, an industrial-grade interface, and can be mounted on a DIN35 rail. The industrial control computer can not only control the walking mechanism, lifting mechanism, and camera components, water quality sensor components 11, and light source 12 included in the detection platform, but also transmit underwater fish images and water color images collected by the camera components to the fish disease identification intelligent subsystem through a wireless router.

[0028] exist Figure 1 The intelligent inspection device includes a housing 1, a walking mechanism mounted on the housing 1 and slidably connected to a closed track 4 set on the aquaculture facility, a lifting mechanism 7 mounted on the side of the housing 1 away from the walking mechanism, and a detection platform mounted on the lifting mechanism 7 via a first support frame. The closed track 4 enables the lifting mechanism 7 to operate automatically along the track on the aquaculture facility. When it reaches a designated position on the aquaculture facility, it automatically extends and retracts to extend the detection platform into the designated position in the aquaculture water to obtain information on fish activity. The housing 1 is also equipped with a UWB positioning tag 5 and an alarm light 6.

[0029] exist Figure 1 In this structure, the walking mechanism includes a walking wheel 2 slidably connected to a closed track 4 installed on the aquaculture facility and a drive assembly 3 for driving the walking wheel 2 to rotate. The drive assembly 3 includes a drive motor and a reducer located inside the housing 1. The output shaft of the drive motor is connected to the walking wheel 2 via a transmission shaft. It should be noted that the walking mechanism is powered by the drive assembly 3, which drives the walking wheel 2 to roll on the closed track 4, thereby enabling the walking mechanism to reciprocate along the length of the track. The lifting mechanism 7 moves synchronously with the walking mechanism, thereby enabling the lifting mechanism 7 to move horizontally along the closed track 4. The walking mechanism ensures stable operation along the track and does not deviate or derail through the limiting and guiding structure, achieving precise positioning and movement.

[0030] exist Figure 1 The detection platform includes a second support frame 8 connected to the lifting mechanism 7 via a first support frame, and a camera assembly, a water quality sensor assembly 11, and a light source 12 mounted on the second support frame 8. The second support frame 8 is a stainless steel bracket for mounting the camera assembly, water quality sensor assembly 11, and light source 12, which keeps the camera of the camera assembly at a certain distance and angle from the light source 12, and can adjust the immersion depth according to the water depth and fix it in a designated position.

[0031] The camera assembly includes a first camera 9 for acquiring images of fish at different water layers within the aquaculture facility and a second camera 10 for acquiring water color images and spectral data within the aquaculture facility. In this embodiment, the first camera 9 is an underwater industrial camera, and the second camera 10 is a multispectral camera. The angles of the first camera 9, the second camera 10, and the light source 12 can all be adjusted by changing the installation position of the positioning bolts. The light source 12 is a 30-50W underwater low color temperature LED light source or an infrared light source, which can automatically or manually supplement and adjust the light according to the underwater lighting conditions. The underwater industrial camera is waterproof, has a 360° adjustable angle, 4K resolution, and a frame rate ≥30fps, and can clearly capture lesion areas as small as 0.5 mm.

[0032] The second embodiment of the present invention relates to an intelligent inspection method for aquaculture fish diseases, applied to a fish disease identification intelligent agent subsystem. This subsystem includes a fish disease identification model, a water quality-linked diagnostic module, and an incremental learning framework. The process is as follows: Figure 3 As shown, the details are as follows:

[0033] Step 101: Acquire water color images and spectral data of the aquaculture facility collected by the second camera 10, and preprocess the water color images and spectral data of the aquaculture facility.

[0034] Step 102: Extract features from the preprocessed water color image and spectral data, and input the features into the water quality AI inversion model to obtain water quality parameters for the aquaculture facility.

[0035] Step 103: Compare water quality parameters with preset fishery water quality standard thresholds, and determine anomalies and risk levels of the water environment in the aquaculture facility based on the comparison results.

[0036] Specifically, steps 101 to 103 correspond to the water quality linkage diagnosis module. This module can determine multiple parameters such as dissolved oxygen, turbidity, pH, and ammonia nitrogen through water color. It also includes a trained water quality AI inversion model. Water color images and spectral data are collected by the second camera 10, a multispectral camera. After preprocessing and feature extraction, these images are input into the trained water quality AI inversion model to quantitatively invert water quality parameters such as dissolved oxygen, pH, and ammonia nitrogen based on water color features. This enables non-contact, multi-parameter synchronous and rapid diagnosis. The identified water quality parameters are then compared with preset fishery water quality standard values ​​to determine anomalies and classify risks.

[0037] Step 104: Acquire fish images at different water levels within the aquaculture facility captured by the first camera 9, along with water quality parameters and aquaculture logs for the aquaculture facility, and preprocess the fish images at different water levels within the aquaculture facility.

[0038] Step 105: Extract fish body features from the preprocessed fish body images, and obtain multimodal features based on the fish body features extracted from the fish body images, water quality parameters, and aquaculture logs.

[0039] Specifically, lightweight depthwise separable convolution and improved channel attention mechanism are used to extract fish body features from the preprocessed fish body images, and water quality parameters and features from the aquaculture log are also extracted. The fish body features are then fused with the features from the water quality parameters and the aquaculture log to obtain multimodal features.

[0040] Step 106: Input the multimodal features into the fish disease identification model, and obtain the species identification and disease diagnosis of fish in the aquaculture facility through the fish disease identification model.

[0041] Specifically, steps 104 to 105 correspond to the fish disease identification model; the fish disease identification agent first collects multimodal data such as fish images, water quality environment, and aquaculture logs, then preprocesses the fish images and uses lightweight depthwise separable convolution (…). ) and improved channel attention mechanism ( Fish physical characteristics are then combined with features extracted from water quality and aquaculture logs to obtain multimodal features through hierarchical fusion. The fish disease identification model employs an improved CIoU loss (…). Complete fish body localization, lesion detection, and identification of vital signs changes, combined with cross-entropy loss ( It enables category identification and disease diagnosis, and is based on a comprehensive health index ( Complete health assessments, conduct etiological analysis through rule-based reasoning, and ultimately automatically output disease diagnoses, etiological conclusions, and standardized prevention and treatment strategies.

[0042] Step 107: Based on the identification of fish species and diagnosis of diseases in the aquaculture facility, as well as the judgment of abnormalities and risk classification of the water quality environment in the aquaculture facility, automatically match the corresponding regulatory agents and treatment methods from the preset drug library, drug use rule library, and prevention and control plan library, and calculate the drug concentration and dosage of the corresponding regulatory agents.

[0043] Specifically, this invention matches fish species identification results, health assessment levels, disease diagnosis conclusions, and etiology analysis results from a pre-set drug library, medication rule library, and prevention and control plan library; automatically calculates the type, concentration, and dosage of medication based on aquaculture water volume, stocking density, water temperature, dissolved oxygen, pH value, and ammonia nitrogen parameters; generates medication recommendations including drug name, dosage, usage method, timing of use, and precautions, following the logic of emergency treatment first, then disease treatment, and finally environmental control and prevention; simultaneously, it generates a comprehensive prevention and control plan including water quality control, feeding management, oxygenation strategies, and disease prevention, based on the causes of water quality abnormalities, and outputs the medication recommendations and prevention and control plan for execution.

[0044] The main interface of this invention includes functional modules such as system on-site pictures, monitoring data, report list, equipment monitoring historical data, and alarm information; the inspection report supports PDF export or printing; this invention integrates diverse equipment such as underwater inspection cameras and mobile handheld terminals, and is suitable for various aquaculture scenarios such as factory farming and cage farming.

[0045] Furthermore, the incremental learning framework in this invention includes a data increment module, a model increment module, a knowledge retention module, and an evaluation and optimization module. The data increment module continuously feeds in new fish disease sample data; the model increment module updates model parameters based on improved deep learning algorithms (such as lightweight convolution + attention mechanism); the knowledge retention module prevents the model from forgetting historical knowledge through methods such as distillation loss; and the evaluation and optimization module verifies model performance and adjusts parameters in real time, thereby optimizing the identification model. The optimization process is as follows: first, the data increment module collects new fish disease sample data and preprocesses it; then, the model increment module uses the new samples to fine-tune network parameters; combined with the knowledge retention module, historical diagnostic knowledge is retained; finally, the evaluation and optimization module verifies the model accuracy and iteratively adjusts it until it meets the requirements, achieving accurate identification of new fish disease types and new lesion characteristics, and improving the model's adaptability and robustness.

[0046] In practical applications, the specific embodiments of the present invention will be described in detail below in the context of factory farming. Please refer to the following for details. Figure 4 and Figure 5 :

[0047] 1. System Installation: A closed track 4 is installed on the factory-style aquaculture pond. The walking mechanism is mounted on the track. The detection platform is connected to the lifting mechanism 7 through the first support frame. By adjusting the angle of the camera and the light source 12, the image information of fish at different water layers in the aquaculture pond can be clearly collected. A small waterproof industrial control computer is installed in the control cabinet next to the aquaculture pond. It is connected to the cloud server through a wireless bridge and configured with edge computing nodes.

[0048] 2. Data Acquisition: After the system is started, the walking mechanism drives the lifting mechanism 7 to move along the fixed track. After reaching the designated breeding area, the lifting mechanism 7 automatically extends and retracts, sending the detection platform into the target water layer; the light source 12 automatically supplements the light according to the underwater illuminance, and the camera of the waterproof camera collects underwater fish body images and water color images, and transmits the image data to the industrial control computer.

[0049] 3. Disease Diagnosis: The industrial control computer transmits the collected underwater fish images and water color images to the fish disease identification intelligent agent. Based on a database and improved deep learning algorithms, and combined with parameters such as dissolved oxygen, pH, and ammonia nitrogen provided by the water quality linkage diagnosis module, the intelligent agent completes fish species identification, health assessment, disease diagnosis, and etiology analysis. The entire identification response process is less than 3 seconds, and the detection accuracy is less than 0.5 mm in lesion areas. The specific steps are as follows: First, high-definition images of various parts of the fish body are acquired through the multimodal data acquisition module, focusing on capturing skin erythema, ulceration, head congestion, deformities, and fin... The images are preprocessed to remove underwater environmental interference by denoising and enhancing lesion features such as broken scales, congestion, scale loss, and defects. Then, an improved lightweight CNN combined with an improved channel attention mechanism is used to extract lesion features from each part and enhance the signal of small lesions. Next, a multimodal feature fusion module is used to improve the robustness of recognition by combining data such as water quality and aquaculture logs. Finally, the fish disease recognition model is used to accurately classify and locate gill rot, parasitic infection (such as white spots and parasite attachment), and water mold disease (such as flocculent hyphae) based on the lesion feature database.

[0050] 3. Prevention and Control Plan Generation: The system automatically generates medication recommendations and prevention and control plans based on the diagnostic results, and displays alarm information on the main interface; farmers can view the diagnostic report through mobile handheld terminals, and the report supports PDF export or printing.

[0051] 4. Model Optimization: During system operation, new disease data and diagnostic results are continuously collected through an incremental learning framework to optimize the fish disease identification model and improve the accuracy of subsequent diagnoses.

[0052] In summary, the fish disease identification intelligent agent in this invention is based on an improved deep learning algorithm, combined with computer vision and multimodal data fusion technology. Through a fish disease identification model, it is used to identify farmed species and changes in vital signs, and provides health assessment, disease diagnosis, etiology analysis, and prevention and control strategies. The fish disease identification intelligent agent supports accurate identification of major aquatic diseases such as gill rot, parasitic infections, and saprolegniasis on the skin, head, fins, and scales. It can detect lesions as small as 0.5 mm with a response time of less than 3 seconds. By employing an improved CIoU loss function (\(L_{CIoU} = 1 - IoU+ \frac{\rho^2(b,b^{gt})}{c^2} + \alpha v\)) to optimize bounding box regression, the accuracy of small target localization is improved. Combined with image super-resolution enhancement technology, the details of small lesions are magnified, achieving detection of lesion areas larger than 0.5 mm. Through depthwise separable convolution (\(D_{c} = I_{c} * W_{c}\) and \(O = The lightweight network structure (\sum_{c=1}^{C} D_{c} * W_{1×1,c}) reduces computational load and optimizes the model inference process, enabling rapid feature extraction and matching with a recognition response time of less than 3 seconds, ensuring that the response speed meets the actual aquaculture needs.

[0053] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the invention. All equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in this invention should still be covered by the claims of this invention.

Claims

1. An intelligent inspection system for farmed fish diseases, characterized in that: This includes a fish disease information collection device and a fish disease identification intelligent subsystem connected to it via the Internet; The fish disease information collection device includes an intelligent inspection device located in the aquaculture facility and a data transmission device and an industrial control computer located in the control cabinet. The industrial control computer is connected to the intelligent inspection device and the fish disease identification intelligent body subsystem through the data transmission device. The intelligent inspection device includes a housing (1), a walking mechanism installed on the housing (1) and slidably connected to a closed track (4) set on the aquaculture facility, a lifting mechanism (7) installed on the side of the housing (1) away from the walking mechanism, and a detection platform installed on the lifting mechanism (7) through a first support frame.

2. The intelligent fish disease inspection system for aquaculture according to claim 1, characterized in that: The walking mechanism includes a walking wheel (2) that is slidably connected to a closed track (4) set on the aquaculture facility and a drive assembly (3) for driving the walking wheel (2) to rotate. The drive assembly (3) includes a drive motor and a reducer located in the housing (1). The output shaft of the drive motor is connected to the walking wheel (2) through a transmission shaft.

3. The intelligent inspection system for aquaculture diseases according to claim 1, characterized in that: The detection platform includes a second support frame (8) connected to the lifting mechanism (7) via a first support frame, and a camera assembly, a water quality sensor assembly (11), and a light source (12) mounted on the second support frame (8). The camera assembly includes a first camera (9) for acquiring images of fish at different water layers in the aquaculture facility and a second camera (10) for acquiring water color images and spectral data of the water body in the aquaculture facility.

4. A method for intelligent inspection of diseases in farmed fish, characterized in that: The intelligent inspection system for farmed fish diseases based on any one of claims 1-3, and applied to a fish disease identification intelligent agent subsystem, wherein the fish disease identification intelligent agent subsystem includes a fish disease identification model, a water quality linkage diagnosis module, and an incremental learning framework, and the method includes the following steps: Acquire fish images of different water layers in the aquaculture facility collected by the first camera (9), as well as water quality parameters and aquaculture logs of the aquaculture facility, and preprocess the fish images of different water layers in the aquaculture facility; Fish body features are extracted from the preprocessed fish body images, and multimodal features are obtained based on the fish body features extracted from the fish body images, water quality parameters, and aquaculture logs. The multimodal features are input into the fish disease identification model, and the fish disease identification model is used to obtain the species identification and disease diagnosis of the fish in the aquaculture facility.

5. The intelligent inspection method for farmed fish diseases according to claim 4, characterized in that: Before acquiring images of fish at different water layers within the aquaculture facility captured by the first camera (9), as well as water quality parameters and aquaculture logs of the aquaculture facility, and before preprocessing the images of fish at different water layers within the aquaculture facility, the method further includes: Acquire water color images and spectral data of the aquaculture facility collected by the second camera (10), and preprocess the water color images and spectral data of the aquaculture facility; Features are extracted from the preprocessed water color image and spectral data, and the features are input into the water quality AI inversion model to obtain water quality parameters for aquaculture facilities. The water quality parameters are compared with the preset fishery water quality standard thresholds, and the water quality environment in the aquaculture facility is judged to be abnormal and risk-classified based on the comparison results.

6. The intelligent inspection method for farmed fish diseases according to claim 4, characterized in that: The process of extracting fish body features from preprocessed fish images and obtaining multimodal features based on these features, water quality parameters, and aquaculture logs includes: In the preprocessed fish images, lightweight depthwise separable convolution and improved channel attention mechanism are used to extract fish features, and the features in the water quality parameters and aquaculture logs are also extracted. The fish features are then fused with the features in the water quality parameters and aquaculture logs to obtain multimodal features.

7. The intelligent inspection method for farmed fish diseases according to claim 4, characterized in that: The process of obtaining species identification and disease diagnosis of fish within the aquaculture facility through the fish disease identification model includes: The fish disease identification model uses an improved CIoU loss to locate fish bodies, detect lesions, and identify changes in vital signs. It also combines cross-entropy loss to identify the species of fish and diagnose diseases within the aquaculture facility.

8. The intelligent inspection method for farmed fish diseases according to claim 5, characterized in that: After inputting the multimodal features into the fish disease identification model and obtaining the species identification and disease diagnosis of the fish in the aquaculture facility through the fish disease identification model, the method further includes: Based on the identification of fish species and diagnosis of diseases in the aquaculture facility, as well as the judgment of abnormalities and risk classification of the water quality environment in the aquaculture facility, the corresponding regulatory agents and treatment methods are automatically matched from the preset drug library, drug use rule library, and prevention and control plan library, and the drug concentration and dosage of the corresponding regulatory agents are calculated.

9. The intelligent inspection method for farmed fish diseases according to claim 4, characterized in that: It also includes the following steps: New fish disease sample data is collected through an incremental learning framework. The network parameters in the fish disease identification model are fine-tuned using the new fish disease sample data and combined with historical diagnostic knowledge to obtain an optimized fish disease identification model. The accuracy of the optimized fish disease identification model is verified, and iterative adjustments are made until the accuracy of the fish disease identification model meets the requirements, thereby improving the adaptability and robustness of the fish disease identification model.

Images