Fish disease monitoring method and system based on feature image recognition

Abstract

The application relates to the field of data processing, in particular to a fish disease monitoring method and system based on feature image recognition. The method comprises the following steps: monitoring the water quality change in a target area in real time through a water quality sensor; collecting fish image information in the target area; performing behavior recognition on the fish image information to obtain fish behavior information of each fish object in the target area; performing lesion recognition on the fish behavior information, the fish image information and the water quality change through a fish lesion recognition model to obtain potential lesion information of each fish object; and generating lesion risk information of each fish object according to the potential lesion information and displaying the lesion risk information in a visual model corresponding to the target water area. The method can realize individualized potential fish lesion recognition for different types of fish, improve the accuracy of fish lesion monitoring, simplify the fish lesion monitoring process and further improve the monitoring efficiency of fish lesions.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of data processing, and in particular to a fish disease monitoring method and system based on feature image recognition. BACKGROUND

[0002] Fish live in water, and the water provides a certain concealment for them, which makes it difficult for monitoring personnel to directly observe the activities and health status of fish. Especially for fish in some large water bodies such as lakes and oceans, the monitoring difficulty is greater. Moreover, the behaviors of fish in water are complex and diverse, and the behavior habits of different types of fish are quite different. Some fish may have nocturnal habits or prefer to live in difficult-to-observe locations such as the bottom of the water, which increases the difficulty of observing their behaviors comprehensively and accurately, and behavior abnormalities are often an important signal of disease occurrence.

[0003] Therefore, there is an urgent need for a new solution to overcome the above problems and achieve automatic monitoring of fish diseases. SUMMARY

[0004] The present application provides a fish disease monitoring method and system based on feature image recognition to realize individualized potential fish disease identification for different types of fish, improve the accuracy of fish disease monitoring, simplify the fish disease monitoring process, and further improve the monitoring efficiency of fish diseases.

[0005] In a first aspect, the present application provides a fish disease monitoring method based on feature image recognition, comprising:

[0006] monitoring the water quality change in the target area in real time through a water quality sensor; the water quality change includes the following parameters: chlorophyll concentration, dissolved oxygen, temperature, and pH value;

[0007] collecting fish image information in the target area;

[0008] performing behavior recognition on the fish image information to obtain fish behavior information of each fish object in the target area;

[0009] performing disease identification on the fish behavior information, the fish image information, and the water quality change through a fish disease identification model to obtain potential disease information of each fish object;

[0010] generating disease risk information of each fish object according to the potential disease information and displaying it in the visual model corresponding to the target water area.

[0011] In a second aspect, the present application provides a fish disease monitoring system based on feature image recognition, which comprises the following units:

[0012]

[0012] a monitoring unit configured to monitor water quality changes in the target area in real time by using a water quality sensor, wherein the water quality changes include the following parameters: chlorophyll concentration, dissolved oxygen, temperature, and pH value; and collect image information of fish in the target area;

[0013] an identification unit configured to identify behaviors of the fish in the image information to obtain fish behavior information of each fish object in the target area, and identify potential disease information of each fish object by using a fish disease identification model based on the fish behavior information, the image information of the fish, and the water quality changes;

[0014] a display unit configured to generate disease risk information of each fish object based on the potential disease information, and display the disease risk information in a visual model corresponding to the target water area.

[0015] In a third aspect, an electronic device is provided, and the electronic device includes:

[0016] at least one processor, a memory, and an input / output unit;

[0017] The memory is configured to store a computer program, and the processor is configured to invoke the computer program stored in the memory to execute the fish disease monitoring method based on feature image recognition of the first aspect.

[0018] In a fourth aspect, a computer-readable storage medium is provided, and the computer-readable storage medium includes instructions that, when executed on a computer, cause the computer to execute the fish disease monitoring method based on feature image recognition of the first aspect.

[0019] The fish disease monitoring method and system based on feature image recognition have the following advantages: the water quality changes in the target area are monitored in real time by using a water quality sensor; the water quality changes include the following parameters: chlorophyll concentration, dissolved oxygen, temperature, and pH value; the image information of fish in the target area is collected; the behaviors of the fish in the image information are identified to obtain fish behavior information of each fish object in the target area; the fish behavior information, the image information of the fish, and the water quality changes are identified by using a fish disease identification model to obtain potential disease information of each fish object; the disease risk information of each fish object is generated based on the potential disease information, and the disease risk information is displayed in a visual model corresponding to the target water area. The technical solution can realize individualized potential fish disease identification for different types of fish, improve the accuracy of fish disease monitoring, simplify the fish disease monitoring process, and further improve the monitoring efficiency of fish disease. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1is a flowchart of a fish disease monitoring method based on feature image recognition according to an embodiment of the present application;

[0021] Figure 2 is a structural diagram of a fish disease monitoring system based on feature image recognition according to an embodiment of the present application;

[0022] Figure 3 is a structural diagram of an electronic device according to an embodiment of the present application;

[0023] Figure 4 is a structural diagram of a medium device according to an embodiment of the present application. DETAILED DESCRIPTION

[0024] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0025] The embodiments of the present application provide a fish disease monitoring method and system based on feature image recognition.

[0026] In the embodiments of the present application, initial image data to be processed is acquired; the initial image data contains a target object to be analyzed in an industrial monitoring scene; the initial image data is three-dimensionally spliced to obtain a spliced image corresponding to the target object; virtual modeling is performed based on the spliced image to obtain an initial virtual model of the target object; the virtual initial model is a virtual three-dimensional model of the target object in a spliced image space; the initial virtual model is spatially morphologically optimized through an adaptive optimization model to obtain an optimized virtual model; the optimized virtual model is a virtual three-dimensional model of the target object in an image optimization space; and three-dimensional point cloud data of the target object is extracted from the optimized virtual model to complete intelligent extraction of the three-dimensional point cloud data.

[0027] The embodiment of the present application, one, through the water quality sensor real-time monitoring water quality change situation, can obtain the dynamic information of water quality in the target area in time, such as the change of chlorophyll concentration, dissolved oxygen, temperature, pH value and other parameters. At the same time, fish image information is collected and behavior recognition is carried out, and the behavior state of fish can also be mastered in real time. This real-time monitoring can find abnormalities in the early stage of disease occurrence, which provides the possibility for timely prevention and control measures, avoids the further spread and diffusion of disease. Compared with the traditional single method relying on naked eye observation or part parameter judgment, the fish disease identification model can comprehensively analyze and identify the potential disease information of fish by comprehensively analyzing and identifying the fish behavior information, fish image information and water quality change. Multi-dimensional data fusion analysis can improve the accuracy of identification and reduce the possibility of misjudgment and omission.

[0028] Second, not only pay attention to the image and behavior information of fish, but also include water quality change in the monitoring range. Water quality is an important factor affecting fish health, and many diseases are closely related to water quality changes. By comprehensively considering these factors, the environment and health status of fish can be more comprehensively understood, and whether the fish has disease risk can be judged from multiple angles. The potential disease information and disease risk information of each fish object in the target area can be obtained by individual disease identification and risk assessment. This individual-level analysis can more accurately locate the problem, help to treat and manage the diseased fish, and also better monitor the health status of the whole fish population.

[0029] Third, the disease risk information of each fish object generated according to the potential disease information is displayed to the visualization model corresponding to the target water area, which is presented to the user in an intuitive way. Users can clearly understand the health status distribution of the fish population through the visualization interface, quickly locate the areas and individuals with disease risk, and take timely measures. It provides a scientific basis for decision-making for breeders or relevant management personnel. They can develop reasonable breeding strategies according to the visual disease risk information, such as adjusting water quality, optimizing feed feeding, isolating diseased fish in time, etc., so as to effectively prevent and control the occurrence and spread of fish diseases, improve breeding efficiency and reduce economic losses.

[0030] Fourthly, through real-time monitoring and comprehensive analysis, potential disease risks can be detected before the fish diseases become obvious, achieving early warning. This enables the aquaculture personnel to take preventive measures in advance, such as strengthening water quality management, increasing fish immunity, etc., to control the disease in the embryonic state and reduce the possibility and harm of disease outbreak. The large amount of monitoring data accumulated by the system can be used to analyze the trend of fish health status and the occurrence of diseases. Through the mining and analysis of these data, the aquaculture personnel can better understand the occurrence mechanism and influencing factors of fish diseases, so as to develop more forward-looking disease prevention and control plans, optimize the aquaculture environment and management measures, and improve the overall health level of fish.

[0031] Fifthly, when collecting fish image information and monitoring water quality changes, there is no need to directly contact the fish, avoiding the stress and harm caused to the fish by catching or sampling operations. This non-contact monitoring method helps to maintain the natural living state of the fish, reduces the changes in the health status of the fish caused by human interference, and improves the authenticity and reliability of the monitoring results. Compared with some traditional disease detection methods (such as dissection), the system does not cause damage to the fish, does not affect the growth and development of the fish, and does not destroy the overall ecological environment of the fish population. This is of great significance for protecting fish resources and maintaining the balance of the aquaculture ecosystem.

[0032] In summary, in the embodiments of the present application, individualized potential fish disease identification is realized for different types of fish, improving the accuracy of fish disease monitoring, simplifying the fish disease monitoring process, and further improving the monitoring efficiency of fish diseases

[0033] The fish disease monitoring scheme based on feature image recognition provided in the embodiments of the present application can also be executed by an electronic device, which can be a server, a server cluster, a cloud server. The electronic device can also be a terminal device such as a mobile phone, a computer, a tablet computer, a wearable device, or a special-purpose device (such as a special-purpose terminal device with a fish disease monitoring system based on feature image recognition, etc.). The above-mentioned chips introduced in the embodiments can also be carried in these electronic devices. Alternatively, these electronic devices can also install a service program for executing the fish disease monitoring scheme based on feature image recognition.

[0034] Figure 1 A schematic diagram of a fish disease monitoring method based on feature image recognition provided in the embodiments of the present application is shown in Figure 1 The method comprises the following steps:

[0035] 101. Real-time monitoring of water quality changes in the target area by a water quality sensor;

[0036] 102. Collecting fish image information in the target area;

[0037] 103. performing behavior recognition on the fish image information to obtain fish behavior information of each fish object in the target region;

[0038] 104. performing lesion recognition on the fish behavior information, the fish image information, and the water quality change condition by a fish lesion recognition model to obtain potential lesion information of each fish object;

[0039] 105. generating lesion risk information of each fish object according to the potential lesion information and displaying the lesion risk information into a visual model corresponding to the target water area.

[0040] In the embodiments of the present application, the water quality change condition includes the following parameters: chlorophyll concentration, dissolved oxygen, temperature, and pH value. The chlorophyll concentration mainly reflects the number and growth condition of phytoplankton (algae, etc.) in the water body. In a pond for breeding crucian carp, if the chlorophyll concentration suddenly increases, for example, from 10 μg / L to 30 μg / L, it may mean that the algae in the water body are proliferating in large quantities. On the one hand, the photosynthesis of algae can produce oxygen and increase the dissolved oxygen content of the water body. On the other hand, if the algae proliferate excessively, it may cause the transparency of the water body to decrease, affecting the penetration of light, and when the algae die in large quantities, they consume oxygen in the water and produce some harmful substances at the same time, which threatens the living environment of the crucian carp and increases the risk of disease, such as gill disease, etc. Conversely, if the chlorophyll concentration is too low, for example, lower than 5 μg / L, it may mean that the water body is too barren and lacks phytoplankton, which affects the growth and nutrient intake of the crucian carp.

[0041] Dissolved oxygen is an essential condition for fish survival. Taking the water body for breeding tilapia as an example, the suitable dissolved oxygen content for the growth of tilapia is generally above 4 mg / L. When the dissolved oxygen content decreases to 3 mg / L, the tilapia will show a decrease in appetite and slow growth; if it continues to decrease to below 2 mg / L, the tilapia may float its head and even suffocate to death in severe cases. For example, during the high-temperature period in summer, due to the vigorous activity of microorganisms in the water, the oxygen consumption increases, and if oxygen is not increased in time, the dissolved oxygen in the water body may decrease, causing the tilapia to lack oxygen, decrease resistance, and be susceptible to various diseases, such as gill rot disease, red skin disease, etc. When the dissolved oxygen content is sufficient, for example, reaches 5 mg / L or above, the tilapia has good feeding and growth conditions and relatively strong resistance to diseases.

[0042] Temperature has a significant impact on the physiological activities of fish, and different species of fish have their own suitable temperature range for survival. For example, rainbow trout is a cold-water fish, and its suitable growth temperature is generally between 12-18℃. If the water temperature rises above 20℃, the metabolism of rainbow trout will accelerate, the oxygen consumption will increase, and its immunity may decrease, making it vulnerable to bacterial, viral and other pathogens, leading to diseases such as infectious hematopoietic organ necrosis disease. On the contrary, when the water temperature is too low, such as below 8℃, the appetite of rainbow trout will decrease, and the growth rate will slow down. Long-term exposure to low temperature environment may also cause frostbite, and further cause diseases such as water mold disease.

[0043] pH value reflects the acidity and alkalinity of the water body. In the pond where South American white shrimp is cultured, the suitable pH value range of South American white shrimp is generally between 7.8-8.6. If the pH value is too high, reaching 9.0 or above, the water body is too alkaline, which can corrode the gill tissue of shrimp, affect its respiratory function, and also may enhance the toxicity of some substances in the water body, causing slow growth, difficulty in molting, and even death of shrimp. If the pH value is too low, such as below 7.0, the water body is too acidic, which can affect the absorption of minerals such as calcium by shrimp, leading to soft shell and decreased resistance, and easily infected with bacterial diseases such as black gill disease.

[0044] These water quality parameters are interrelated and influence each other, and jointly act on the survival and health of fish. By monitoring these parameters in real time, the breeding environment can be adjusted in time to prevent the occurrence of fish diseases.

[0045] In step 101, it is assumed that the target area is a large freshwater pond where a large number of grass carp are cultured. Water quality sensors are installed in the pond, which can monitor parameters such as chlorophyll concentration, dissolved oxygen, temperature, pH value, etc. in real time.

[0046] For example, during the summer high temperature period, the sensor detects that the water temperature gradually rises to above 30℃ (exceeding the upper limit of the suitable growth water temperature of 20-30℃ for grass carp), and the dissolved oxygen content decreases from the normal 6mg / L to below 4mg / L.

[0047] Therefore, through the above step 101, water quality change information can be obtained in real time, providing environmental data basis for subsequent analysis of fish health status. Cultivation personnel can adjust the water quality in time according to these data, such as starting the oxygenation equipment to increase the dissolved oxygen, to avoid fish stress or disease caused by deteriorating water quality.

[0048] In step 102, a high-definition camera is installed above the pond to take pictures of the grass carp in the pond at regular intervals. The camera can cover most of the area of the pond to ensure that fish in different positions can be photographed.

[0049] For example, a high-definition picture is taken every 15 minutes, and the picture clearly shows the size, color, and state of the grass carp.

[0050] Thus, through the above step 102, the intuitive visual information of the fish can be obtained, providing image data for subsequent behavior recognition and lesion recognition. These image information can be one of the important bases for judging the health status of the fish.

[0051] In step 103, the fish images taken are analyzed by computer vision and machine learning algorithms.

[0052] For example, the algorithm identifies the swimming speed, swimming direction, and gathering state of the grass carp. If it is found that the originally dispersed grass carp begins to gather in a corner of the fish pond, or the swimming speed of individual grass carp is significantly slower than other fish.

[0053] In this way, the behavior characteristics of the fish are extracted from the images, and these behavior information can reflect the health and physiological state of the fish. Abnormal behavior may be an early signal of fish disease, which helps to discover potential health problems in time.

[0054] In step 104, it is assumed that the fish lesion recognition model is a deep learning model trained by a large amount of data. Based on this assumption, the fish behavior information (such as swimming abnormalities) obtained in the previous step, the fish image information (such as the appearance of spots on the body surface), and the water quality changes (such as high water temperature and low dissolved oxygen) are input into the model. The model analyzes these multi-dimensional data and determines that a grass carp may have gill rot disease (because of the deterioration of water quality, the abnormal behavior of the fish and the possible slight damage to the body surface, etc.).

[0055] By integrating multiple aspects of information for lesion recognition, the accuracy and reliability of the recognition are improved. No longer relying on a single piece of information to determine whether the fish is sick, it can more comprehensively analyze the health status of the fish and discover potential lesion risks.

[0056] In step 105, the potential lesion information of each grass carp determined by the model is quantitatively processed to generate lesion risk information.

[0057] For example, for the grass carp that may have gill rot disease, a higher lesion risk level (such as 80% likelihood of disease) is given, and this information is displayed in a visualization model with a fish pond as the background, and the positions of the fish with high risk of disease are marked with different colors or icons.

[0058] In this way, the health status of the fish population is intuitively displayed to the aquaculture personnel, facilitating the aquaculture personnel to quickly understand which fish have a higher risk of disease, so that targeted measures can be taken, such as isolating sick fish, treating them, etc., improving the efficiency and effectiveness of aquaculture management, and reducing economic losses.

[0059] As an optional embodiment, in 103, the fish image information is subjected to behavior recognition to obtain fish behavior information of each fish object in the target region, including:

[0060] A local feature image containing the to-be-recognized object in the fish image information is extracted; the local feature image is compared with standard fish feature images in a fish variety database to obtain a fish variety to which the to-be-recognized object belongs; and based on the fish variety, the fish image information is subjected to behavior recognition to obtain a fish behavior type contained in the fish image information and corresponding fish behavior parameters as the fish behavior information.

[0061] For example, in an actual fish breeding environment, such as a large pond in which multiple fish species are mixed, the fish image information collected by a camera often contains numerous fish individuals and surrounding environmental information. In order to accurately recognize the behavior of each fish, a local feature image containing the to-be-recognized object (i.e., a single fish) needs to be extracted from the overall image first. This can use target detection algorithms in computer vision, such as Faster R-CNN, YOLO, etc. based on deep learning, which can accurately locate fish individuals in the image and segment them from the background and other fish to obtain a local feature image containing only a single fish. For example, for a carp swimming in the pond, the algorithm can accurately frame the carp and extract its local image for subsequent analysis.

[0062] A rich fish variety database is established, which contains standard feature images of various common breeding fish species, recording the appearance characteristics of different fish varieties in a normal state, such as body shape, color, fin shape and position, etc. The local feature image of the to-be-recognized object extracted is compared with the standard images in the database. Through image matching algorithms, such as SIFT (Scale-Invariant Feature Transform), SURF (Speeded-Up Robust Features) algorithms based on feature points, or image classification algorithms based on deep learning, the similarity of the to-be-recognized image and each standard image in the database is calculated, and the standard image with the highest similarity is found to determine the fish variety to which the to-be-recognized object belongs. For example, for the extracted local feature image, after comparison with the database, it is determined that the fish is a grass carp.

[0063] Different species of fish have different behavior patterns and habits. After determining the species of fish, the image information of the fish is recognized based on the behavior characteristics of the species. Using a behavior recognition model trained specifically for this species of fish (which can be a model based on machine learning or deep learning), the fish's actions, postures, and other information are analyzed to determine its behavior type, such as swimming, feeding, resting, and avoiding. At the same time, further parameters related to these behaviors can be extracted, such as swimming speed, swimming direction, feeding frequency, etc. For example, for the identified grass carp, the model can determine that it is currently performing a feeding behavior, and can calculate its feeding frequency as 3 times per minute, swimming speed as 10 cm per second, etc. These behavior types and parameters together constitute the fish behavior information of the grass carp.

[0064] By first determining the species of fish and then recognizing behavior based on species characteristics, the behavior differences between different species of fish can be fully considered, avoiding confusion between the behavior patterns of different species of fish, thereby improving the accuracy of behavior recognition. For example, the swimming methods and speeds of carp and crucian carp may be different, and accurate identification of the species can more accurately determine whether their behavior is normal. In actual fish farming, mixed farming is more common. This method can accurately identify the behavior of each species of fish in a complex mixed farming environment, providing detailed fish school behavior information for farmers, which helps better manage and monitor the health status of fish schools in mixed farming ponds. For example, in a pond that raises grass carp, carp, and crucian carp, the behavior of different species of fish can be monitored and analyzed. Not only can the behavior type of the fish be identified, but also detailed parameters related to it can be obtained, providing more valuable information for in-depth study of the physiological state and health status of the fish. Farmers can discover abnormal behavior of the fish in a timely manner based on these parameters, such as a sudden decrease in feeding frequency, which may indicate a problem with the health of the fish, so that appropriate measures can be taken for prevention and treatment. Based on accurate fish species identification and behavior information, farmers can develop individualized farming management strategies for different species of fish, such as adjusting the amount of feed, optimizing the farming environment, etc., to improve farming efficiency and economic benefits, while also helping to ensure the healthy growth of fish.

[0065] Further optionally, the fish behavior parameters are determined based on the fish behavior type. If the fish behavior type includes movement behavior, the fish behavior parameters at least include: fish movement distance, movement direction, movement time, movement trajectory type. If the fish behavior type includes predation, the fish behavior parameters at least include: prey object type, predation time, predation frequency, predation frequency. If the fish behavior type includes hibernation, the fish behavior parameters at least include: hibernation time, hibernation frequency, hibernation location.

[0066] Specifically, the following introduces the above fish behavior parameters respectively in combination with specific examples.

[0067] Fish moving distance: Assuming in a bass breeding pond, an image recognition system monitors a bass moving from one side to the other side of the pond within a period of time. By using the scale or other reference information in the image, combined with image recognition algorithm to determine the position of the bass at different times, the moving distance can be calculated. For example, after calculation, this bass moves about 5 meters in 10 minutes. The breeder can analyze the moving distance to judge the activity range and vitality of the bass. If the moving distance suddenly decreases, it may mean that the health of the bass has problems, or the environmental factors such as water quality have changed.

[0068] Moving direction: Similarly for the above bass, the image recognition system can determine its moving direction according to its position change in different frames of images. For example, it is found that this bass moves from the northwest to the southeast of the pond. Understanding the moving direction of fish can help the breeder understand the activity pattern of the fish school, for example, some fish may have fixed migration direction at certain time period, if the direction is abnormal, it may indicate that there are external interference factors, such as water pollution, water flow change, etc.

[0069] Moving time: Record the time points when the bass starts and ends moving, so as to obtain its moving time. For example, this bass starts moving at 10:05 am and stops at 10:15 am, the moving time is 10 minutes. The analysis of moving time can be combined with other parameters to judge the activity intensity and behavior pattern of fish. For example, if the moving time of bass decreases significantly during the day and increases at night, it may indicate that its adaptation to environmental factors such as light has changed.

[0070] Moving trajectory type: By continuously recording the position information of the bass at different times, its moving trajectory can be drawn. The moving trajectory type may include straight line type, curve type, spiral type, etc. For example, it is observed that the moving trajectory of this bass presents irregular curve type, which may reflect its behavior of searching for food, exploring environment or avoiding predators, etc. Different moving trajectory types can provide clues for the breeder about the behavior motivation and environmental adaptability of fish.

[0071] Prey type: In a salmon farm pond, an image recognition system identifies salmon feeding. Further image analysis can determine the type of prey, such as small shrimp, plankton, or other fish. For example, it was found that this salmon was feeding on a type of small shrimp in the pond. Understanding prey type is crucial for fish farmers to properly formulate feeds and manage the aquaculture ecosystem. If salmon primarily prey on a specific organism, and that organism is insufficient, it may be necessary to adjust the feed formulation to meet the salmon's nutritional needs.

[0072] Predation time: Record the times when the salmon begins and ends its predation to determine the predation period. For example, if this salmon begins predating at 2:30 PM and ends at 2:40 PM, the predation time is 10 minutes. The length of the predation time can reflect the salmon's predation efficiency and hunger level. If the predation time is too long, it may mean that the number of prey is scarce or that the salmon's predation ability has decreased, requiring further observation and analysis.

[0073] Feeding frequency: This counts the number of times a salmon feeds within a specific time period. For example, if a salmon feeds 5 times in one hour, changes in feeding frequency can serve as an indicator of the salmon's health and appetite. A sudden decrease in feeding frequency may indicate that the salmon is sick or uninterested in its current food.

[0074] Predation frequency: The predation frequency is calculated based on the number and timing of predations. For example, if the salmon predates 5 times in one hour, the predation frequency is once every 12 minutes. Analyzing the predation frequency can help fish farmers understand the feeding patterns of salmon, rationally arrange the timing and amount of feed, and improve feed utilization.

[0075] Hibernation Time: Taking farmed eels as an example, an image recognition system monitors whether the eel has entered a hibernation state. The start and end times of hibernation are recorded to determine the hibernation time. For example, if this eel begins hibernation at 10 PM and wakes up at 6 AM the next morning, the hibernation time is 8 hours. The length of the hibernation time has a significant impact on the eel's growth and health. If the hibernation time is too short or too long, it may indicate a problem with the eel's physiological state, requiring further investigation of the farming environment and the eel's health condition.

[0076] Hibernation frequency: This counts the number of times an eel hibernates within a certain period. For example, if this eel hibernates 10 times in a week, the frequency can reflect its living habits and environmental adaptability. A sudden increase or decrease in hibernation frequency may be related to changes in environmental factors such as water temperature and water quality.

[0077] Hibernation site: Determine the specific location where eels hibernate, such as a corner of the pond, among the water plants, or in a cave. Understanding the hibernation site of eels can help aquaculture personnel optimize the breeding environment and provide more suitable hibernation sites. For example, if eels often choose to hibernate in a certain corner, aquaculture personnel can increase some shelter in that area to improve the safety and comfort of eels.

[0078] As an optional embodiment, after comparing the local feature image with the standard fish feature image in the fish variety database in 103 to obtain the fish variety to which the to-be-identified object belongs, the climate conditions of the current period can also be monitored in real time. Then, the water area type where each fish object appears and the geographical location range of the target area are obtained. Next, based on the climate conditions, the water area type, and the geographical location range of the target area, the matching degree between the fish variety and the target area is verified. If the matching degree is not lower than the set threshold, the verified fish variety is taken as the final output target fish variety.

[0079] After identifying the fish variety, real-time acquisition of the climate conditions of the current period is performed by using meteorological monitoring equipment (such as a weather station, satellite cloud image data, etc.). The climate conditions include multiple key factors, such as temperature, humidity, light intensity, air pressure, wind force, and wind direction. Taking a large fish breeding pond located on the sea coast as an example, on a summer day, the weather station monitors that the air temperature reaches 32°C, the humidity is 70%, the light intensity is strong, the wind force is level 3, and the wind direction is southeast. These climate condition information is crucial for subsequent judgment of the matching degree between the fish variety and the target area.

[0080] The water area type where the fish is located is determined by means of field investigation, geographic information system (GIS), etc. The water area type is rich and diverse, including but not limited to freshwater lakes, rivers, seawater breeding ponds, offshore sea areas, reservoirs, etc. For example, it is monitored that the target area is a small freshwater reservoir located in a mountainous area, which has obvious differences in water quality, flow speed, water depth, etc. from other water area types.

[0081] The geographical location range of the target area is determined by means of GPS positioning technology, map data, etc. It can be accurate to the latitude and longitude coordinates, or it can be described by a more macro geographical region, such as a specific province, city, and village, etc. For example, the target area is located in a coastal city in Guangdong Province in southern China, and the climate, hydrology, and other geographical environments of this area have unique characteristics.

[0082] Each fish has its suitable living climate conditions, preferred water area type, and specific geographical distribution range. The suitable living conditions of the identified fish variety are compared and analyzed with the actual situation monitored at present, and the matching degree between them is calculated.

[0083] For example, the identified fish species is salmon, which is generally suitable for living in cold water environment with low water temperature and clear water quality, mainly distributed in cold water sea areas of high latitude or cold water streams of mountainous areas. The current target area is a mariculture pond located in a tropical region, with high water temperature in summer, and the water quality and flow conditions are quite different from the suitable living environment of salmon. Through comprehensive analysis of the climate conditions (high temperature), water area type (tropical sea pond), and geographical location range (tropical region), it is concluded that the matching degree of salmon and the target area is low.

[0084] The calculation of matching degree can adopt the way of setting multiple index weights and scoring, such as climate conditions accounting for 40% weight, water area type accounting for 30% weight, and geographical location range accounting for 30% weight. According to the compliance degree of each index, the corresponding score is given, and finally the total matching degree is calculated by weighted calculation.

[0085] A matching degree threshold (such as 70%) is set. If the calculated matching degree of the fish species and the target area is not less than the threshold, it means that the fish species is reasonable to survive in the current target area, and it is determined as the final target fish species. For example, the identified fish species is tilapia, which is verified to be suitable for living in warm water. The target area is a freshwater aquaculture pond in a city in southern China. The current climate is warm, and the water area type and geographical location range are matched with the suitable living conditions of tilapia, with a matching degree of 80%, which is higher than the set threshold of 70%. Therefore, the final target fish species is determined as tilapia.

[0086] Therefore, by combining factors such as climate conditions, water body area types, and geographical location ranges, the identified fish species are verified, avoiding potential misjudgments that may occur solely relying on image feature recognition. For example, some fish species may have similar appearances to other species during their juvenile stages or when influenced by the environment. Through environmental factor verification, the species can be more accurately determined, reducing subsequent management problems caused by false identification. Accurate fish species information is crucial for breeding decisions. After understanding the matching degree of fish species to the target area, breeders can adjust breeding strategies based on the suitable living conditions of fish, such as water quality adjustment, feed feeding, and breeding density control. For example, if it is determined that the fish species is suitable for cold water environments, but the current target area water temperature is higher, breeders can take measures such as increasing water change frequency, building sunshades, etc. to improve the success rate of breeding and economic benefits. Ensuring that the introduced fish species matches the target area can effectively prevent ecological risks caused by the introduction of alien species or the inability to adapt to the environment. If fish species that are not suitable for the local environment are introduced, it may cause damage to the local ecosystem, such as competing for resources with local species, spreading diseases, etc. By verifying the matching degree, such situations can be avoided, protecting the balance and stability of the ecological environment. After determining the accurate fish species and suitable breeding environment, breeding resources can be more reasonably utilized, avoiding waste of resources. For example, based on the habits and needs of fish, precise feeding and reasonable arrangement of breeding facilities can improve resource utilization efficiency and reduce breeding costs.

[0087] As an optional embodiment, in 103, if there are multiple fish species with matching degrees higher than the set threshold, the fish species with the highest probability of appearing in the target area is selected from the verified multiple fish species as the target fish species. If the matching degree is lower than the set threshold, the step of comparing the local feature image with the standard fish feature image in the fish species database to obtain the fish species to which the object to be identified belongs is re-executed.

[0088] After calculating the matching degree of the identified fish species with the current target area's climate conditions, water body area types, and geographical location ranges, it may occur that the matching degrees of multiple fish species are all higher than the set threshold. For example, in a freshwater lake breeding area located in the south, several possible fish species are initially determined through image recognition, and after matching degree calculation, it is found that the matching degrees of crucian carp, carp, and grass carp are all higher than the set threshold (assuming 70%). This means that these three species of fish may theoretically survive in this area and have good adaptability to the current environmental conditions.

[0089] When multiple fish species with high matching degrees appear, it is necessary to further determine the most likely target fish species. This can be done by referring to historical data, local fishery statistics, expert experience, etc. to evaluate the probability of each fish species appearing in the target area. For example, by checking local fishery records and sharing experiences of breeders, it is found that the number of carp breeding and the frequency of natural occurrence in the freshwater lake breeding area are relatively higher than those of carp and grass carp, that is, the probability of carp appearing in this area is the highest. Therefore, in this case, carp is determined as the target fish species.

[0090] If the matching degree of all the initially identified fish species with the target area is below the set threshold value after calculation, it means that the currently identified fish species does not match the environmental conditions of the area, and there may be an identification error. For example, in a marine aquaculture area, the matching degrees of several fish species identified initially are all below 70% after matching degree calculation. This indicates that these fish species are unlikely to survive in this marine area, and the current identification result may not be accurate.

[0091] When the matching degree is below the set threshold value, in order to obtain more accurate fish species identification results, the system will re-compare the extracted local feature images with the standard fish feature images in the fish species database. By using image comparison algorithms such as feature point matching algorithm, deep learning image classification algorithm, etc. again, the fish species is re-identified in order to find a fish species that matches the environmental conditions of the target area better.

[0092] From the technical effect, when multiple fish species with high matching degree appear, by selecting the fish species with the highest probability of appearing in the target area, the range can be further narrowed down, and the accuracy of the finally determined fish species can be improved. This avoids the uncertainty caused by multiple species meeting the basic conditions, making the recognition result more in line with the actual situation. When the matching degree is low, the identification step is re-performed, which can timely correct possible misidentification, continuously optimize the identification result, and ensure that the finally determined fish species is adapted to the environment of the target area. This processing method enables the system to cope with different situations, whether there are multiple possible fish species or the identification result does not match the environment. It can be adjusted and optimized through corresponding operations. This improves the adaptability and reliability of the system in complex environments, and better meets the demand for accurate identification of fish species in actual breeding process. The target fish species obtained through the above steps is more accurate, and the breeding personnel can develop more reasonable breeding strategies based on the characteristics and needs of the species, such as selecting appropriate feed, controlling breeding density, adjusting water quality management measures, etc. This helps to improve breeding efficiency, reduce breeding risk, and increase breeding income. Accurate identification of fish species and ensuring its match with the target area helps to rationally plan and utilize fishery resources. It avoids introducing fish species that are not suitable for the local environment, reduces resource waste and ecological risk, protects the local fishery ecological balance, and promotes the sustainable development of fishery resources.

[0093] As an optional embodiment, the fish disease identification model at least includes the following structures: an extraction layer, a prediction layer, and a correction layer.

[0094] Based on the above model, in 104, the fish disease identification model is used to identify the fish behavior information, the fish image information, and the water quality change to obtain potential disease information of each fish object, including:

[0095] 201. The fish behavior information, the fish image information, and the water quality change are extracted by the extraction layer to obtain image feature information of each fish object. The image feature information of each fish object at least includes fish behavior features, surface image features, and water quality change features of each fish object.

[0096] 202. The prediction layer is used to predict the surface disease probability of each fish object based on the surface image features and water quality change features of each fish object.

[0097] 203. The correction layer is used to correct the surface disease probability of each fish object by using the fish behavior features of each fish object to obtain the potential disease information of each fish object.

[0098] In step 201, fish behavior information, fish image information, and water quality change are taken as inputs. These information reflect the living state of fish from different angles, for example, fish behavior information can reflect its daily activity pattern, fish image information can directly show the surface condition, and water quality change reflects the environmental factors of fish survival.

[0099] The input multi-source data is feature extracted by related algorithms or models in the extraction layer. For fish behavior information, features such as moving speed, activity frequency, and social behavior may be extracted; for fish image information, surface image features such as color, texture, and shape may be extracted, such as whether there are spots, ulcers, and fin state; for water quality change, features of parameters such as chlorophyll concentration, dissolved oxygen, temperature, and pH value may be extracted. Finally, image feature information containing fish behavior features, surface image features, and water quality change features is obtained, providing a basis for subsequent analysis.

[0100] In step 202, the surface image features and water quality change features obtained by the extraction layer are input into the prediction layer. This is because fish surface lesions are often closely related to the appearance changes of the surface and the water quality environment for survival. For example, too low dissolved oxygen in water may cause fish to have difficulty breathing, which in turn affects its health, and some abnormal symptoms may appear on the surface; and features such as spots and ulcers in the surface image are directly related to lesions.

[0101] The prediction layer uses existing models and algorithms to predict the surface lesion probability of each fish object according to the input feature information. This process may be based on machine learning, deep learning, etc., through learning and analysis of a large amount of historical data to establish a relationship model between features and lesion probability. For example, if irregular white spots appear in the surface image features and the water quality change features show that the water temperature abnormally rises, the model may predict that the fish has a high surface lesion probability.

[0102] In step 203, the fish behavior features obtained by the extraction layer are introduced into the correction layer. Fish behavior changes are also an important basis for judging its health status. For example, when fish are sick, they may exhibit abnormal behavior, such as slow swimming, swimming alone, losing balance, etc.

[0103] The correction layer corrects the surface lesion probability obtained by the prediction layer according to the fish behavior features. If the fish behavior features show obvious abnormalities, such as long periods of inactivity or frequent jumping, even if the surface lesion probability given by the prediction layer is low, the correction layer will appropriately increase the probability to more accurately reflect the potential lesion condition of the fish. Conversely, if the fish behavior is normal and the probability given by the prediction layer is high, the correction layer may reduce the probability to avoid misjudgment.

[0104] Thus, by comprehensively considering fish behavior information, fish image information, and water quality changes, the health status of fish is evaluated from multiple dimensions, avoiding inaccurate judgments that may be caused by relying on a single information source. For example, only through the body surface image, some temporary changes in water quality may cause misjudgment of body surface abnormalities as lesions, while combining the characteristics of water quality changes and fish behavior can more accurately identify the true lesion condition. The model can discover potential problems in the early stages of fish lesions. Even if the fish body surface has not yet appeared obvious symptoms of lesions, but there have been some abnormal behaviors, or the water quality environment has changed, which is not conducive to the health of fish, the model can analyze this information to predict and correct the lesion probability in advance, providing valuable time for timely prevention and control measures. Accurate lesion identification results can help aquaculture personnel develop more scientific aquaculture management strategies. If the model finds that some fish have a high probability of potential lesions, aquaculture personnel can adjust the water quality, increase the nutrition of feed, isolate and treat sick fish, etc., thereby effectively reducing the incidence and mortality of fish diseases and improving aquaculture efficiency. The multi-layer structure of the model and the use of multi-source data make it adaptable to different aquaculture environments and fish species. Whether in pond culture, reservoir culture, or seawater culture, whether for common fish species or some special species, the model can achieve relatively accurate lesion identification through learning and analysis of the corresponding data, and has strong generalization ability and adaptability.

[0105] Further optionally, in 202, according to the body surface image features and water quality change features of each fish object, the body surface lesion probability of each fish object is predicted, including:

[0106] According to the water quality change features of each fish object, the visibility level in the target area is determined; based on the visibility level, the body surface image features of each fish object are sharpened to obtain the optimized body surface image features of each fish object; multi-dimensional visual feature extraction is performed on the optimized body surface image features to obtain the multi-dimensional body surface image features of each fish object; the multi-dimensional body surface image features at least include: body surface texture features, body surface brightness features, and body surface color features; based on a pre-set body surface lesion standard image library, the correlation between the multi-dimensional body surface image features and the body surface lesion standard image is predicted, and the body surface lesion probability of each fish object is calculated based on the correlation.

[0107] Specifically, some parameters in the water quality change feature, such as chlorophyll concentration and suspended substance content, will directly affect the transparency of the water body, and further affect the visibility in the target area. For example, when the chlorophyll concentration is high, the water body may appear a dense green color, resulting in reduced visibility; and the increase of suspended substance content will also make the water body turbid, and the visibility will be poor. By establishing a relevant mathematical model or empirical formula, these parameters in the water quality change feature are associated with the visibility level. For example, set the chlorophyll concentration in a certain range to correspond to the low visibility level, in another range to correspond to the medium visibility level, and so on. In this way, according to the water quality change feature of the region where each fish object is located, the visibility level of the region can be determined.

[0108] Different visibility levels will affect the clarity of the fish body surface image taken. For the case of low visibility level, the image may be relatively blurred, and the details are difficult to identify. Therefore, according to the determined visibility level, the corresponding image sharpening algorithm is used to process the body surface image feature. For example, for the image of low visibility level, Laplace sharpening operator or Gaussian sharpening filter, etc. can be used to enhance the edges and details of the image, so that the texture, spots and other features of the fish body surface are more clear. After sharpening, the optimized body surface image feature is more conducive to subsequent analysis.

[0109] The optimized body surface image feature contains rich information, and these information is further mined through multi-dimensional visual feature extraction method. Among them, the body surface texture feature can reflect the surface structure of the fish body surface, such as the arrangement of scales, whether there are protrusions or depressions, etc.; the body surface brightness feature can reflect the light and dark degree of different parts of the fish body surface, and some diseases may cause changes in local brightness; the body surface color feature can show the color distribution and tone of the fish body surface, and some diseases may change the color of the fish body surface. Using special image analysis algorithms, such as texture analysis algorithm (such as gray level co-occurrence matrix), brightness and color space conversion algorithm, etc., these multi-dimensional body surface image features are extracted from the optimized body surface image feature.

[0110] A large number of fish body surface images with clear lesion types and degrees are collected and sorted in advance to construct a standard image library of body surface lesions. The multi-dimensional body surface image features of each fish object extracted are compared and analyzed with the image features in the standard image library. The similarity or correlation index (such as Euclidean distance, cosine similarity, etc.) between the multi-dimensional body surface image features and the standard image features is calculated to evaluate the matching degree between them. The higher the correlation is, the greater the possibility of the fish having the corresponding lesion is. According to the size of the correlation, combined with a certain probability calculation model (such as a probability prediction model based on machine learning), the body surface lesion probability of each fish object is calculated. For example, if the multi-dimensional body surface image features of a certain fish object have a high correlation with the image features of a certain lesion type in the standard image library, it can be predicted that the fish object has a high body surface lesion probability.

[0111] Therefore, by sharpening the body surface image features according to the visibility level, the image blurring problem caused by water quality factors can be effectively improved, making the image clearer, and thus improving the accuracy of subsequent feature extraction. Accurate feature extraction is the basis for lesion prediction, providing more reliable data support for subsequent analysis.

[0112] The multi-dimensional visualization feature extraction method can analyze fish body surface images from multiple angles and fully mine the information in the images. By considering multiple feature dimensions such as body surface texture, brightness, and color, the features of the fish body surface can be described in more detail, and some subtle changes that may be early signs of lesions can be captured, which helps to detect diseases in fish early.

[0113] Based on the standard image library of body surface lesions, correlation analysis and probability calculation are performed to compare the body surface features of the fish with known lesion features, making the prediction results more objective and accurate. By quantifying the correlation to calculate the lesion probability, the error of subjective judgment is avoided, and the reliability of the prediction is improved, providing a scientific basis for aquaculture personnel to take preventive measures in a timely manner.

[0114] The influence of water quality changes on visibility is considered, and the image is processed accordingly, so that the system can adapt to different water quality environments. Whether in clear water or turbid water, accurate fish body surface image features can be obtained through corresponding processing methods, ensuring the stability and reliability of lesion prediction and improving the adaptability and universality of the system.

[0115] In the embodiments of the present application, the calculation process of the correlation between the body surface image feature points Pi in the multi-dimensional body surface image features and the standard body surface lesion images is represented by the following formula:

[0116] ;

[0117] wherein, represents the correlation between the body surface image feature point Pi and the jth body surface lesion standard image, and the weight factor and is used to adjust the contribution degree of the jth body surface lesion standard image in the x dimension and y dimension respectively, x and y represent the projection component coordinate values of the body surface image feature point Pi in the x dimension and y dimension.

[0118] represents the actual coordinate value of the body surface image feature point Pi in the x dimension, represents the corresponding standard coordinate value of the body surface image feature point Pi in the x dimension, represents the average coordinate value of all body surface image feature points in the x dimension in the multi-dimensional body surface image feature, and m is the number of body surface image feature points in the multi-dimensional body surface image feature, is used to represent the average correlation fluctuation range between each body surface image feature point in the multi-dimensional body surface image feature and the jth body surface lesion standard image in the x dimension.

[0119] represents the actual coordinate value of the body surface image feature point Pi in the y dimension, represents the corresponding standard coordinate value of the body surface image feature point Pi in the y dimension, represents the average coordinate value of all body surface image feature points in the y dimension in the multi-dimensional body surface image feature, is used to represent the average correlation fluctuation range between each body surface image feature point in the multi-dimensional body surface image feature and the jth body surface lesion standard image in the y dimension.

[0120] The formula is used to calculate the correlation between the body surface image feature point Pi in the multi-dimensional body surface image feature and the body surface lesion standard image. This formula can comprehensively measure the relationship between the feature point and the standard image from the two-dimensional space by considering the coordinate values and related parameters in the x dimension and y dimension respectively. For example, by calculating and in the x dimension and y dimension respectively, the deviation of the feature point Pi in the respective dimension relative to the average coordinate can be reflected, so as to understand the position characteristics of the feature point in the overall image.

[0121] The setting of the weight factor can flexibly adjust the contribution degree of the weight factor in the x dimension and y dimension to the correlation calculation according to different body surface lesion standard images j. This means that for different types of body surface lesion standard images, the influence of certain dimensions can be emphasized or weakened according to their characteristics and importance, so that the correlation calculation is more flexible and accurate.

[0122] The coordinate value difference considers the difference between the actual coordinate value of the feature point P_i in the x dimension and the y dimension and the standard coordinate value. This difference reflects the similarity in shape between the feature point and the standard image. The smaller the difference, the closer the feature point is to the standard image, and the higher the correlation between them. Conversely, the correlation is lower. In this way, subtle differences between the feature point and the standard image can be accurately captured, allowing for more accurate assessment of their correlation.

[0123] The overall fluctuation range measures the average correlation fluctuation range between the feature points of each body image feature in the multi-dimensional body image and the jth body lesion standard image in the x dimension and the y dimension. They can help assess the dispersion of feature points in different dimensions with the standard image, providing a more comprehensive understanding of the relationship between feature points and standard images. If the fluctuation range is small, it means that the correlation between the feature points and the standard image in that dimension is stable. Conversely, if the fluctuation range is large, it means that the correlation between the feature points and the standard image in that dimension varies greatly, and their relationship needs to be more carefully assessed.

[0124] This formula accurately calculates the correlation between the body image feature points and the body lesion standard image by considering multiple factors, providing accurate data support for subsequent correlation-based calculation of fish body lesion probability, and helping to more accurately identify fish body lesions.

[0125] The embodiments of the present application implement individualized potential fish lesion identification for different types of fish, improve the accuracy of fish lesion monitoring, simplify the fish lesion monitoring process, and further improve the efficiency of fish lesion monitoring.

[0126] Figure 2 A structure diagram of a fish disease monitoring system based on feature image recognition provided by the embodiments of the present application is shown in FIG. 1, which includes the following steps: Figure 2

[0127] The monitoring unit is configured to monitor the water quality changes in the target area in real time through a water quality sensor. The water quality changes include the following parameters: chlorophyll concentration, dissolved oxygen, temperature, and pH value. The fish image information in the target area is collected.

[0128] The recognition unit is configured to perform behavior recognition on the fish image information to obtain fish behavior information of each fish object in the target area. The fish behavior information, the fish image information, and the water quality changes are subjected to lesion recognition through a fish lesion recognition model to obtain potential lesion information of each fish object.

[0129] ​The display unit is configured to generate lesion risk information of each fish object according to the potential lesion information, and display the lesion risk information into a corresponding visual model of the target water area.

[0130] Referring to Figure 3 , Figure 3 An embodiment of an electronic device provided in the embodiment of the present application is shown in the figure. As shown in the figure, the embodiment of the present application provides an electronic device 500, which comprises a memory 510, a processor 520, and a computer program 511 stored in the memory 510 and executable on the processor 520, and the processor 520 implements the foregoing embodiments when executing the computer program 511. Figure 3

[0131] Referring to Figure 4 , Figure 4 An embodiment of a computer readable storage medium provided in the embodiment of the present application is shown in the figure. As shown in the figure, the embodiment provides a computer readable storage medium 600, which stores a computer program 611, and the computer program 611 is executed by a processor to implement the foregoing embodiments. Figure 4

[0132] It should be noted that in the above embodiments, the description of each embodiment has its own emphasis, and the parts not described in detail in a certain embodiment can be referred to the related description of other embodiments.

[0133] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage, etc.) containing computer-usable program code.

[0134] The present application is described with reference to flowcharts and / or block diagrams of the method, device (system), and computer program product according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a general-purpose computer, a special-purpose computer, an embedded computer, or other programmable data processing devices to produce a machine, so that the instructions executed by the computer or other programmable data processing devices produce a device that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in one flow or multiple flows and / or blocks Figure 1 The functions specified in one flow or multiple flows and / or blocks

[0135] ​​These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the Figure 1 function specified in the flow or flows and / or blocks Figure 1 of the block or blocks.

[0136] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions that are executed on the computer or other programmable apparatus provide steps for implementing the Figure 1 function specified in the flow or flows and / or blocks Figure 1 of the block or blocks.

[0137] While the preferred embodiments of the application have been described, additional variations and modifications can be made to the embodiments by those skilled in the art once they learn of the basic inventive concepts. Therefore, the appended claims are intended to cover all such additional variations and modifications as fall within the scope of the application.

[0138] Obviously, numerous modifications and variations of the present application are possible in light of the above teachings. It is therefore to be understood that within the scope of the appended claims and their equivalents, the application can be practiced otherwise than as specifically described.

Claims

1. A fish disease monitoring method based on feature image recognition, characterized by, The method includes at least: The water quality changes in the target area are monitored in real time using water quality sensors; the water quality changes include the following parameters: chlorophyll concentration, dissolved oxygen, temperature, and pH value. Collect image information of fish within the target area; Behavior recognition is performed on the fish image information to obtain fish behavior information of each fish object within the target area; By using a fish lesion identification model, the fish behavior information, the fish image information, and the water quality changes are used to identify lesions in order to obtain potential lesion information for each fish species. Based on the potential lesion information, lesion risk information for each fish species is generated and displayed in the visualization model corresponding to the target water area; The fish lesion recognition model includes at least the following structure: extraction layer, prediction layer, and correction layer; The method involves using a fish lesion identification model to identify lesions from fish behavioral information, fish image information, and water quality changes, in order to obtain potential lesion information for each fish species. The extraction layer extracts features from the fish behavior information, the fish image information, and the water quality changes to obtain image feature information for each fish object. The image feature information for each fish object includes at least the following: fish behavior features, body surface image features, and water quality change features. The prediction layer predicts the probability of skin lesions on the body surface of each fish species based on the surface image features and water quality change features. By using the correction layer, the fish behavior characteristics of each fish species are adopted to correct the probability of lesions on the body surface of each fish species, so as to obtain the potential lesion information of each fish species. The method of predicting the probability of surface lesions for each fish species based on their surface image features and water quality change characteristics includes: Based on the water quality change characteristics of each fish species, the visibility level in the target area is determined; Based on the visibility level, the surface image features of each fish object are sharpened to obtain the optimized surface image features of each fish object. Multidimensional visualization feature extraction is performed on the optimized body surface image features to obtain multidimensional body surface image features for each fish object; the multidimensional body surface image features include at least: body surface texture features, body surface brightness features, and body surface color features; Based on a pre-set standard image library of body surface lesions, the correlation between the multidimensional body surface image features and the standard images of body surface lesions is predicted, and the probability of body surface lesions for each fish species is calculated based on the correlation.

2. The fish disease monitoring method based on feature image recognition according to claim 1, characterized in that, The step of performing behavior recognition on the fish image information to obtain fish behavior information of each fish object within the target area includes: Extract local feature images of the object to be identified from the fish image information; The local feature image is compared with standard fish feature images in the fish species database to determine the fish species to which the object to be identified belongs. Based on the fish species, the fish image information is used to perform behavior recognition to obtain the fish behavior type and corresponding fish behavior parameters contained in the fish image information, which are used as the fish behavior information. 3.The fish disease monitoring method based on feature image recognition according to claim 2, characterized in that, The fish behavior parameter is determined based on the fish behavior type; If the fish behavior type includes movement behavior, the fish behavior parameter at least includes fish movement distance, movement direction, movement time, and movement trajectory type; If the fish behavior type includes predation, the fish behavior parameter at least includes predation object type, predation time, predation frequency, and predation frequency; If the fish behavior type includes dormancy, the fish behavior parameter at least includes dormancy time, dormancy frequency, and dormancy location. 4.The fish disease monitoring method based on feature image recognition according to claim 2, characterized in that, After the comparison of the local feature image with the standard fish feature image in the fish breed database to obtain the fish breed to which the to-be-identified object belongs, the following steps are further included: Real-time monitoring of the climate condition in the current period; Obtaining the water area type in which each fish object appears and the geographical location range of the target area; Verifying the matching degree between the fish breed and the target area based on the climate condition, the water area type, and the geographical location range of the target area; If the matching degree is not lower than a set threshold, the verified fish breed is taken as the final output target fish breed. 5.The fish disease monitoring method based on feature image recognition according to claim 4, characterized in that, After the verification of the matching degree between the fish breed and the target area based on the climate condition, the water area type, and the geographical location range of the target area, the following steps are further included: If there are multiple fish breeds with a matching degree higher than the set threshold, the fish breed with the highest appearance probability in the target area is selected from the multiple verified fish breeds as the target fish breed; If the matching degree is lower than the set threshold, the step of comparing the local feature image with the standard fish feature image in the fish breed database to obtain the fish breed to which the to-be-identified object belongs is re-executed. 6.The fish disease monitoring method based on feature image recognition according to claim 1, wherein, The calculation process of the correlation between the body surface image feature point Pi in the multi-dimensional body surface image feature and the body surface lesion standard image is represented by the following formula: ； wherein, represents the correlation between the feature point Pi of the body surface image and the jth body surface lesion standard image, and the weight factor and λ i2 are used to adjust the contribution degree of the jth body surface lesion standard image in the x dimension and the y dimension, respectively, and x and y represent the projection component coordinate values of the feature point Pi of the body surface image in the x dimension and the y dimension. represents the actual coordinate value of the body surface image feature point P; in the xth dimension, represents the corresponding standard coordinate value of the body surface image feature point P; in the xth dimension, represents the average coordinate value of all body surface image feature points in the xth dimension in the multi-dimensional body surface image feature, and m is the number of body surface image feature points in the multi-dimensional body surface image feature, is used to represent the average correlation fluctuation range between the xth dimension of each body surface image feature point in the multi-dimensional body surface image feature and the jth body surface lesion standard image. represents the actual coordinate value of the body surface image feature point P; in the yth dimension, represents the corresponding standard coordinate value of the body surface image feature point P; in the yth dimension, represents the average coordinate value of all body surface image feature points in the yth dimension in the multi-dimensional body surface image feature, is used to represent the average correlation fluctuation range between each body surface image feature point in the yth dimension and the jth body surface lesion standard image in the multi-dimensional body surface image feature.

7. A fish disease monitoring system based on feature image recognition, characterized by, The system at least includes the following units: A monitoring unit for real-time monitoring of water quality changes in the target area through a water quality sensor; the water quality changes include the following parameters: chlorophyll concentration, dissolved oxygen, temperature, and pH value; and fish image information in the target area is collected; An identification unit for behavior identification of the fish image information to obtain fish behavior information of each fish object in the target area; A fish lesion identification model is used to identify the fish behavior information, the fish image information, and the water quality changes to obtain potential lesion information of each fish object; A display unit is used to generate lesion risk information of each fish object according to the potential lesion information and display it in the visual model corresponding to the target water area; The fish lesion identification model at least includes the following structures: an extraction layer, a prediction layer, and a correction layer; The fish lesion identification model at least includes the following structures: an extraction layer, a prediction layer, and a correction layer; The fish behavior information, the fish image information, and the water quality change are extracted by an extraction layer to obtain image feature information of each fish object, wherein the image feature information of each fish object at least includes fish behavior features, body surface image features, and water quality change features of each fish object; The body surface disease probability of each fish object is predicted by a prediction layer according to the body surface image features and the water quality change features of each fish object; The body surface disease probability of each fish object is corrected by a correction layer using the fish behavior features of each fish object to obtain potential disease information of each fish object; The body surface disease probability of each fish object is predicted according to the body surface image features and the water quality change features of each fish object, including: The visibility level in the target area is determined according to the water quality change features of each fish object; The body surface image features of each fish object are sharpened based on the visibility level to obtain optimized body surface image features of each fish object; Multi-dimensional visual feature extraction is performed on the optimized body surface image features to obtain multi-dimensional body surface image features of each fish object, wherein the multi-dimensional body surface image features at least include body surface texture features, body surface brightness features, and body surface color features; The correlation between the multi-dimensional body surface image features and body surface disease standard images is predicted based on a pre-set body surface disease standard image library, and the body surface disease probability of each fish object is calculated based on the correlation.

8. An electronic device, comprising: The memory is used to store computer software programs; The processor is used to read and execute the computer software programs to realize the fish disease monitoring method based on feature image recognition according to any one of claims 1-5.

Images