Intelligent fishing system and method based on image recognition

Abstract

The invention relates to an intelligent fishing system and method based on image recognition. The system comprises an underwater image acquisition module, an environment monitoring module, an image preprocessing module, an intelligent recognition analysis module, a central control module, a fishing execution module, a wireless communication module and a power supply module. According to the invention, through multi-source data fusion acquisition, adaptive image optimization processing, deep learning-driven accurate fish identification and trajectory prediction, dynamic fishing strategy generation and automatic execution, real-time accurate determination of fish types, sizes and densities and selective fishing are realized; the technical problems that traditional fishing depends on artificial experience, efficiency is low, excessive fishing or missing fishing is prone to being caused, and an existing intelligent fishing technology is poor in adaptability to the complex underwater environment, insufficient in recognition precision and incapable of achieving refined selective fishing are solved, and the method has the advantages of being high in recognition accuracy, high in environmental adaptability, good in fishing controllability and excellent in sustainability.

Description

Technical Field

[0001] This invention relates to the field of intelligent fishing technology, specifically to an intelligent fishing system and method based on image recognition. Background Technology

[0002] Aquaculture, as an important component of my country's fisheries, relies on large-scale and intelligent development to improve aquaculture efficiency and ensure the supply of aquatic products. Fishing operations are one of the core elements of aquaculture. Traditional fishing methods rely heavily on human experience to determine the timing and area of ​​fishing, using fixed-mesh nets for large-scale harvesting. However, human experience is insufficient to accurately determine the species, size, and density of fish in the water, leading to situations where larger fish are caught while smaller ones are missed or juvenile fish are accidentally caught. This not only affects aquaculture efficiency but also undermines the sustainability of the aquaculture ecosystem. The complex and variable underwater environment, such as turbid water, insufficient light, and fluctuating water currents, makes visual identification difficult. Existing identification technologies that rely solely on cameras are prone to significant drops in accuracy. The fishing process requires manual deployment and retrieval of nets, which is labor-intensive, inefficient, and lacks real-time monitoring and dynamic adjustment capabilities. Furthermore, there is no effective linkage between the fishing results and previous monitoring data, making it impossible to optimize subsequent identification and fishing strategies based on actual fishing feedback, resulting in slow technological iteration. Existing intelligent fishing technologies employ sonar to detect fish distribution, but sonar can only obtain the approximate location and density of fish, failing to achieve precise identification of species and size. Other technologies utilize conventional image recognition, but these are not optimized for the underwater environment, resulting in insufficient recognition accuracy, and they are not integrated with adjustable fishing equipment for selective fishing. Therefore, developing an image recognition-based intelligent fishing system and method has become an urgent need for the intelligent development of aquaculture. Summary of the Invention

[0003] (a) Technical problems to be solved To address the shortcomings of existing technologies, this invention provides an intelligent fishing system and method based on image recognition. This system offers advantages such as adaptability to complex underwater environments, accurate fish identification, and selective automated fishing. It enables precise identification and trajectory prediction of underwater fish species, sizes, and densities, unaffected by environmental factors such as water turbidity and lighting changes. It allows for selective fishing by automatically adjusting net size according to fish size to avoid juvenile fish and protect the aquaculture ecosystem. Furthermore, it increases the automation level of fishing operations, reduces labor costs, and improves fishing efficiency and accuracy. Finally, it establishes a data loop, optimizing the identification model and fishing strategy through fishing feedback, enabling continuous technological iteration.

[0004] (II) Technical Solution To achieve the above objectives, the present invention provides the following technical solution: an intelligent fishing system and method based on image recognition, comprising: The underwater image acquisition module is used to acquire underwater fish image data. The underwater image acquisition module integrates a 1080P high-definition waterproof camera (waterproof rating IP68, underwater working depth ≤50m), an LED adjustable light intensity supplementary light unit (light intensity adjustment range 100-1000lm), and a 200kHz ultrasonic auxiliary imaging unit (detection distance 0.5-10m, resolution ≤2cm). The ultrasonic auxiliary imaging unit is used to supplement fish outline information in turbid waters, and the supplementary light unit can adaptively adjust the brightness according to the ambient light intensity. The environmental monitoring module is used to collect underwater environmental parameters, including a water temperature sensor (measurement range 0-40℃, accuracy ±0.1℃), a water quality sensor (detects pH value 6.5-8.5, turbidity 0-50NTU), a dissolved oxygen sensor (measurement range 0-20mg / L, accuracy ±0.1mg / L), and a water pressure sensor (measurement range 0-5atm). The image preprocessing module is used to perform denoising, dehazing, enhancement, and image-ultrasonic data fusion processing on the acquired underwater fish images. The dehazing process adopts an improved dark channel prior algorithm (introducing an adaptive adjustment mechanism for underwater light attenuation coefficient), and the denoising adopts a 3×3 window mid-range filtering algorithm. The intelligent identification and analysis module has a built-in fish feature library and an improved YOLOv8 deep learning model, which is used to identify fish species, measure size, count numbers, calculate cluster density, and predict movement trajectories from preprocessed image data. The central control module uses an STM32H743 microcontroller as the core control unit. It receives the recognition results from the intelligent recognition and analysis module and the environmental parameters from the environmental monitoring module, and generates a dynamic fishing strategy. The fishing strategy includes the adjustment parameters of the fishing net mesh size, the coordinates of the deployment position, the fishing duration, and the timing of retrieval. The fishing execution module includes an adjustable mesh intelligent fishing net, an underwater positioning unit, a drive unit, and a retrieval unit. The adjustable mesh intelligent fishing net achieves stepless adjustment of the mesh size from 5 to 20 cm by extending and retracting the mesh wires via a motor. The wireless communication module uses a combination of 4G / 5G and underwater acoustic communication to realize data transmission between the central control module and remote terminals and various functional modules, with a transmission delay of ≤1s; The power module provides stable power to each module, including a 12V / 100Ah waterproof lithium battery pack and a 100W solar charging unit. It has overcharge, over-discharge, and short-circuit protection functions, and the battery life is ≥72 hours.

[0005] Furthermore, the improved YOLOv8 deep learning model adds CBAM (Convolutional Block Attention Module) to the backbone to enhance the extraction of key features such as fish fins and scales; and uses BiFPN (Bidirectional Feature Pyramid Network) to optimize multi-scale feature fusion in the neck part, which improves the recognition accuracy of small fish with a body length of <10cm and overlapping fish by ≥15%, with an overall recognition accuracy of ≥95% and a detection speed of ≥30fps. Furthermore, the fish feature database contains standard images and annotation information for more than 30 species of freshwater and saltwater fish. The annotations include fish body length, body width, outline features, growth stage, and typical texture information. It supports incremental training and updates based on fishing feedback data, and the feature matching efficiency is improved by 5-8% after each update. Furthermore, the underwater positioning unit adopts a combination of GPS and underwater acoustic positioning beacons, with a positioning accuracy of ≤1m; the drive unit includes two thrusters with a maximum power of 500W and an attitude adjustment mechanism, which can realize fine-tuning of the fishing net's attitude in the horizontal and vertical directions with an adjustment accuracy of ≤0.5°, ensuring that the net can be stably deployed in an environment with a water flow speed of ≤0.8m / s. Furthermore, the adjustable mesh smart fishing net uses 304 stainless steel wire (1.2mm in diameter) and a waterproof stepper motor drive mechanism, and the unfolded area of ​​the net can be 10-50m². 2 The mesh size can be adjusted as needed within the specified range, with an adjustment accuracy of ≤0.5cm. The mesh surface is coated with a polytetrafluoroethylene anti-corrosion coating, and the underwater service life is ≥3 years. A smart fishing method based on image recognition includes the following steps: S1: Data acquisition starts. The underwater image acquisition module acquires underwater image data at a preset frequency of 5fps, and the ultrasonic-assisted imaging unit acquires contour supplementary data simultaneously. The environmental monitoring module acquires water temperature, water quality, dissolved oxygen and water pressure parameters every 10s. All data are transmitted to the central control module in real time through the wireless communication module and stored in the local database. S2: Image preprocessing. After receiving the data, the image preprocessing module sequentially performs median filtering for noise reduction, improved dark channel prior dehazing, and CLAHE adaptive contrast enhancement (clipping limit 2.0, grid size 8×8). Finally, it generates an enhanced fish image through a weighted fusion algorithm (ultrasonic data weight 0.3, image data weight 0.7). S3: Intelligent Recognition and Analysis. The intelligent recognition and analysis module calls the improved YOLOv8 model and combines it with a fish feature database to achieve species identification. It calculates body length / body width using a pixel calibration method (1 pixel = 0.1 cm conversion relationship) and statistically analyzes data from 10m... 2 The number and density of fish within the coverage area are predicted using the Kalman filter algorithm to determine the fish's movement trajectory in the next 5-10 seconds, with a positioning error of ≤0.5m. S4: Fishing strategy generation. The central control module combines the recognition results with environmental parameters to determine the fishing conditions. If the conditions are met, the net size (target fish body length × 0.6), release location (trajectory prediction endpoint), and fishing duration (20-30 min) are determined. If juvenile fish are present, the net size is increased by 5 cm or the fishing area is adjusted. If the environmental parameters do not meet the standards, the fishing is delayed and an early warning is sent. S5: Automated fishing execution. The drive unit deploys the fishing net according to the positioning information and adjusts it to the preset mesh size. During the fishing process, the number of nets caught and the net's posture data are fed back in real time. The drive unit dynamically adjusts the net's posture according to changes in water flow. S6: Fishing and Recovery and Data Feedback. Recovery is initiated after the preset conditions are met. The actual fishing results are statistically analyzed and compared with the identification results. The identification accuracy is calculated, and the data is uploaded to the remote terminal and the fish feature database and model parameters are updated. Furthermore, the pixel calibration method described in S3 is implemented as follows: a standard calibration plate (50cm×50cm in size, 10cm spacing between calibration points) is fixed in advance within the field of view of the underwater image acquisition module. Through calibration experiments at different water depths (0.5-5m), a nonlinear conversion model between pixels and actual size is established to eliminate underwater refraction errors and ensure that the body length measurement error is ≤0.5cm. Furthermore, the fishing conditions in S4 include: target fish body length ≥ preset commercial size, and aggregation density ≥ 2 fish / m². 3 The water temperature is 4-32℃, dissolved oxygen is ≥5mg / L, and pH is 6.5-8.5. If any environmental parameter exceeds the threshold, the system will delay the harvesting for 1-2 hours and then retest. If the system fails to meet the standards after 3 tests, an emergency warning will be sent and the harvesting operation will be stopped. Furthermore, the net posture adjustment logic in S5 is as follows: real-time water pressure data is collected by a water pressure sensor and the water flow velocity is calculated. When the water flow velocity is 0.3-0.5m / s, the posture adjustment mechanism tilts the net by 5-8°; when the water flow velocity is >0.5m / s, it tilts by 10-15° and increases the mesh size by 1-2cm to prevent the net from deforming or fish from escaping. Furthermore, the data feedback in S6 also includes: storing the feedback data of fishing efficiency (catch per unit time), resource density change (difference in fish density in the area before and after fishing), and mesh size adjustment accuracy in association with environmental parameters to form aquaculture fishing data archive, which supports remote terminal export and analysis, and provides data support for subsequent adjustment of aquaculture density and optimization of feeding strategies.

[0006] Compared with the prior art, the technical solution of this application has the following beneficial effects: 1. This invention effectively solves problems such as underwater turbidity and insufficient light by fusing multi-source data from images and ultrasound, improving preprocessing algorithms, and using deep learning models. It improves the accuracy of fish species identification, reduces body length measurement errors, and significantly enhances adaptability to complex underwater environments. The adjustable mesh design combined with precise identification can dynamically adjust the mesh size according to fish size, effectively avoiding juvenile fish, protecting the aquaculture ecosystem, and conforming to the concept of sustainable aquaculture. At the same time, it improves the purity of commercial fish catches and enhances aquaculture efficiency. It achieves full automation from data collection, identification and analysis, strategy generation to harvesting and recovery, eliminating the need for manual on-site operation, improving harvesting efficiency and significantly reducing labor intensity. The feature library and identification model are dynamically updated through harvesting feedback data, continuously improving identification accuracy and the rationality of harvesting strategies. It can be adapted to various aquaculture scenarios such as ponds, cages, and reservoirs.

[0007] 2. This invention achieves real-time accurate determination and selective fishing of fish species, size, and density through multi-source data fusion acquisition, adaptive image optimization processing, deep learning-driven accurate fish identification and trajectory prediction, dynamic fishing strategy generation, and automated execution. It solves the technical pain points of traditional fishing, which relies on human experience, is inefficient, and is prone to overfishing or underfishing, as well as the poor adaptability of existing intelligent fishing technologies to complex underwater environments, insufficient identification accuracy, and inability to achieve refined selective fishing. It features high identification accuracy, strong environmental adaptability, good fishing controllability, and excellent sustainability, and can be widely applied to various aquaculture scenarios such as pond farming and cage farming. Attached Figure Description

[0008] Figure 1 This is a framework diagram of the intelligent fishing system based on image recognition of the present invention; Figure 2 This is a flowchart of the intelligent fishing method based on image recognition according to the present invention. Detailed Implementation

[0009] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0010] Please see Figure 1-2 The intelligent fishing system and method based on image recognition in this embodiment adopts the following technical solution: (I) Intelligent Fishing System Based on Image Recognition This system achieves intelligent operation across the entire process, from data collection, processing, and identification to fishing execution, through the collaborative work of multiple modules. Specifically, it includes the following functional modules: Underwater Image Acquisition Module: As the core of data acquisition, it integrates a 1080P high-definition waterproof camera (IP68 waterproof rating, underwater working depth ≤50m), an LED adjustable light intensity supplementary lighting unit (light intensity adjustment range 100-1000lm, adaptable to different lighting environments), and a 200kHz ultrasonic auxiliary imaging unit (detection distance 0.5-10m, resolution ≤2cm). The supplementary lighting unit can automatically adjust the brightness according to the ambient light intensity, and the ultrasonic unit provides supplementary data on fish outlines when the water is turbid, overcoming the limitations of simple image recognition.

[0011] Environmental monitoring module: Composed of a water temperature sensor (measurement range 0-40℃, accuracy ±0.1℃), a water quality sensor (detects pH value and turbidity), a dissolved oxygen sensor (measurement range 0-20mg / L, accuracy ±0.1mg / L), and a water pressure sensor, it collects underwater environmental parameters in real time to provide environmental basis for the generation of fishing strategies.

[0012] Image preprocessing module: Addressing the characteristics of underwater images such as high noise levels, strong fog, and low contrast, it employs a processing flow of denoising, defogging, enhancement, and fusion. Noise reduction: Median filtering algorithm is used to effectively remove salt and pepper noise caused by underwater suspended particles; Defogging: Improve the traditional dark channel prior algorithm and introduce an adaptive adjustment mechanism for underwater light attenuation coefficient to solve the problem of poor defogging effect under special underwater lighting conditions; Enhancement: Employs the CLAHE adaptive contrast enhancement algorithm to improve the distinction between fish and the background; Fusion: By combining ultrasonic contour data with image data through a weighted fusion algorithm, the contour details of fish in turbid water are supplemented, thereby improving the robustness of recognition. Intelligent recognition and analysis module: Includes a built-in dynamically updated fish feature library and an improved YOLOv8 deep learning model. Fish Feature Database: Contains standard images of 30+ common farmed fish such as grass carp, common carp, and bass, labeled with information such as body length, body width, outline features, and growth stage, and supports incremental training and updates based on fishing feedback data; Improved YOLOv8 model: Added CBAM (Convolutional Block Attention Module) to the backbone to enhance the extraction of key fish features (such as fins and scales); BiFPN (Bidirectional Feature Pyramid Network) was adopted in the neck to optimize multi-scale feature fusion and improve the recognition accuracy of small fish (body length <10cm) and overlapping fish. The model recognition accuracy is ≥95% and the detection speed is ≥30fps. This module enables fish species identification, body length / width measurement, population statistics, and aggregation density calculation (unit: fish / m²). 3It predicts the movement trajectory of fish in the next 5-10 seconds based on the Kalman filter algorithm, with a positioning error of ≤0.5m. Central Control Module: Utilizing an STM32H743 microcontroller as the core control unit, it integrates data storage, logic operations, and instruction generation functions. By receiving identification results from the intelligent identification and analysis module and parameters from the environmental monitoring module, it establishes a fishing decision-making model. Fishing trigger conditions: Target fish body length ≥ preset marketable size (can be set via remote terminal), and aggregation density ≥ 2 fish / m². 3 The environmental parameters (water temperature 4-32℃, dissolved oxygen ≥5mg / L) meet the requirements; Strategy generation logic: Determine the mesh size based on the fish species (e.g., grass carp commercial sizes correspond to mesh sizes of 8-12cm, juvenile fish to mesh sizes of ≥15cm), plan the release location based on the trajectory prediction results, and adjust the fishing time based on environmental parameters (shorten the fishing time to 15-20min when the water flow is rapid). Fishing execution module: includes an adjustable mesh smart fishing net, an underwater positioning unit, a drive unit, and a recovery unit. Adjustable mesh intelligent fishing net: Utilizing stainless steel wire mesh and a motor-driven mechanism, the mesh size is infinitely adjustable within the range of 5-20cm, with an adjustment accuracy of ≤0.5cm. The unfolded area of ​​the net can be set according to requirements (10-50m²). 2 ); Underwater positioning unit: Combining GPS with underwater acoustic positioning beacons, positioning accuracy ≤1m, real-time feedback of net body position; Drive unit: Includes 2 thrusters (maximum power 500W) and attitude adjustment mechanism to achieve precise deployment and attitude stability of the net; Recovery unit: It adopts a winch and traction rope structure, and the recovery speed is adjustable (0.5-1m / s) to ensure the smooth recovery of the net. Wireless communication module: It adopts a combination of 4G / 5G and underwater acoustic communication to realize data transmission between the central control module and remote terminals and various functional modules. The transmission delay is ≤1s, and it supports remote monitoring and command issuance. Power module: It adopts a 12V / 100Ah waterproof lithium battery pack, with a 100W solar charging unit, and has a battery life of ≥72 hours to meet the needs of long-term operation. It also has overcharge, over-discharge and short circuit protection functions.

[0013] (II) Intelligent Fishing Method Based on Image Recognition This method is based on the above system and the specific steps are as follows: S1: Data Acquisition Start The remote terminal issues a fishing preparation command, and the central control module activates the underwater image acquisition module and the environmental monitoring module. The underwater image acquisition module acquires underwater images at a frequency of 5fps, while the ultrasonic-assisted imaging unit simultaneously acquires contour data; the environmental monitoring module collects environmental parameters every 10 seconds, and all data is transmitted in real time to the central control module via the wireless communication module and stored in the local database. S2: Image Preprocessing The image preprocessing module obtains raw data from the central control module and executes the following steps sequentially: Median filtering for noise reduction: Set the window size to 3×3 to remove suspended particle noise; Improved dark channel prior dehazing: Calculate the dark channel map of underwater images, adaptively adjust atmospheric light value and transmittance, and remove the fog effect caused by water scattering; CLAHE Adaptive Contrast Enhancement: Set the clip limit to 2.0 and the grid size to 8×8 to improve the outline and details of fish. Data fusion: The ultrasonic contour data and the enhanced image are fused with a weight of 0.3:0.7 to generate a clear enhanced image of the fish. S3: Intelligent Recognition and Analysis The intelligent recognition and analysis module calls the improved YOLOv8 model to detect augmented images: Species identification: The fish features in the image are compared with the feature database, and the species determination result is output; Size measurement: The body length and width of the fish are calculated using the pixel calibration method (a conversion relationship of 1 pixel = 0.1 cm is established in advance using a calibration board); Density statistics: Statistical analysis of the image coverage area (preset to 10m) 2 The number of fish within the area is converted into aggregation density; Trajectory prediction: Based on the Kalman filter algorithm, the system takes the position coordinates of the fish in 5 consecutive frames as input and predicts its movement trajectory in the next 5-10 seconds. S4: Fishing Strategy Generation The central control module comprehensively judges the identification results and environmental parameters: If the target fish's body length is greater than or equal to the marketable size and its aggregation density is greater than or equal to 2 fish / m² 3 And if the environmental parameters meet the requirements, the following fishing strategy is generated: mesh size = target fish body length × 0.6 (to ensure that juvenile fish can be exposed), release location = trajectory prediction endpoint area, and fishing duration = 20-30 minutes (dynamically adjusted according to density). If a juvenile fish aggregation density of ≥1 fish / m² is detected 3 It automatically increases the mesh size by 5cm or adjusts the placement location to outside the juvenile fish gathering area; If the environmental parameters do not meet the requirements, an early warning message will be sent to the remote terminal, and the fishing will be delayed for 1-2 hours before retesting. S5: Automated Fishing Execution The central control module sends control commands to the fishing execution module: The drive unit accurately deploys the fishing net to the target area based on the positioning information, and the attitude adjustment mechanism adjusts the net to be horizontally unfolded. The motor drives the wire mesh to extend and retract, adjusting it to the preset mesh size. During the fishing process, the underwater image acquisition module monitors the number of fish caught in the net in real time, and the drive unit adjusts the net's posture according to changes in water flow to ensure the fishing effect. S6: Fishing Recovery and Data Feedback When the fishing time reaches the preset value or the number of nets caught is greater than or equal to the target value (which can be set remotely), the central control module instructs the recovery unit to start the winch and smoothly recover the fishing net. After recovery, the species, quantity, and size distribution of the actual fish caught are statistically analyzed and compared with the identification results to calculate the identification accuracy rate. Information such as comparative data, environmental parameters, and fishing efficiency is uploaded to a remote terminal and used to update the fish feature database, incrementally train the improved YOLOv8 model, and optimize subsequent identification and fishing strategies.

[0014] Example 1: Intelligent Fishing Application in Pond Grass Carp Farming System Deployment: The underwater image acquisition module and environmental monitoring module are fixed in the central area of ​​the pond (water depth 3m), and the fishing execution module is deployed on the bank of the pond. The module is connected to the remote control APP through the wireless communication module. The preset commercial specifications for grass carp are ≥30cm in length and the specifications for juvenile fish are <15cm in length. The target catch is 50 fish. Data Acquisition and Preprocessing: After the system is started, the underwater image acquisition module acquires underwater images of the pond. Due to the slight turbidity of the water, the ultrasonic-assisted imaging unit simultaneously acquires contour data. The image preprocessing module generates clear grass carp images through median filtering for noise reduction, improved dark channel prior dehazing, CLAHE enhancement, and data fusion. Intelligent Recognition and Analysis: The improved YOLOv8 model identified 32 grass carp in the image, with body lengths ranging from 12 to 35 cm and a density of 2.8 fish / m². 3 The system predicts that grass carp will move towards the northeast area of ​​the pond within the next 8 seconds; the environmental monitoring module reports that the water temperature is 25℃ and the dissolved oxygen level is 6.2mg / L, which meets the conditions for harvesting.

[0015] Fishing strategy generation: The central control module determines the mesh size to be 18cm (30cm×0.6), the placement location to be the northeast area of ​​the pond (the predicted trajectory endpoint), and the fishing duration to be 25min. Fishing execution: The propeller of the fishing execution module deploys the fishing net to the target area based on the positioning information, the motor drives the net wire to adjust to 18cm mesh size, and the attitude adjustment mechanism adjusts the net body level according to the water flow; during the fishing process, the number of grass carp entering the net increases in real time. Recovery and Feedback: Recovery was initiated after 25 minutes, with 48 grass carp actually caught, including 46 marketable fish with a body length of ≥30cm and 2 juvenile fish (95% slip rate); the recognition accuracy rate was 93.75%. The data was uploaded to the APP to update the fish feature database and to incrementally train the model. Example 2: Intelligent Fishing Application in Net Cage Bass Farming System Deployment: Two underwater image acquisition modules are installed inside the net cage (10m×10m×5m), the environmental monitoring module is deployed in the corner of the net cage, and the fishing execution module is integrated into the net cage frame; the preset marketable size for bass is ≥25cm in body length, and the target catch density is 3 fish / m³. 3 . Data Acquisition and Preprocessing: The water in the cage is clear. The underwater image acquisition module mainly collects image data, and the preprocessing module generates clear images after noise reduction and enhancement. The environmental monitoring module reports that the water temperature is 22℃ and the dissolved oxygen is 7.0mg / L, which meets the fishing conditions. Intelligent identification and analysis: The model identified 280 bass in the net cage, with 220 of them having a body length ≥25cm, resulting in a density of 4.4 fish / m². 3 It is predicted that the fish will concentrate in the central area of ​​the cages. Fishing strategy generation: Determine the mesh size as 15cm (25cm×0.6), the placement position as the middle of the net cage, and the fishing time as 20min. Fishing execution: The fishing net is precisely deployed to the center of the net cage, the mesh size is adjusted to 15cm, and the propeller is used to help fix the net's posture; the number of nets caught is monitored in real time during the fishing process. Recovery and Feedback: After recovery, 215 bass were actually caught, all of which were marketable fish with a body length of ≥25cm, with an identification accuracy of 97.7%; after data feedback, the model parameters were updated, and the subsequent identification accuracy was further improved.

[0016] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0017] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An intelligent fishing system based on image recognition, characterized in that, include: The underwater image acquisition module is used to acquire underwater fish image data. The underwater image acquisition module integrates a 1080P high-definition waterproof camera, an LED adjustable light intensity supplementary light unit, and a 200kHz ultrasonic auxiliary imaging unit. The ultrasonic auxiliary imaging unit is used to supplement the outline information of fish in turbid waters, and the supplementary light unit can adaptively adjust the brightness according to the ambient light intensity. The environmental monitoring module is used to collect underwater environmental parameters, including water temperature sensor, water quality sensor, dissolved oxygen sensor and water pressure sensor; The image preprocessing module is used to perform denoising, dehazing, enhancement, and image-ultrasonic data fusion processing on the acquired underwater fish images. The dehazing process adopts an improved dark channel prior algorithm, and the denoising process adopts a 3×3 window value filtering algorithm. The intelligent identification and analysis module has a built-in fish feature library and an improved YOLOv8 deep learning model, which is used to identify fish species, measure size, count numbers, calculate cluster density, and predict movement trajectories from preprocessed image data. The central control module uses an STM32H743 microcontroller as the core control unit. It receives the recognition results from the intelligent recognition and analysis module and the environmental parameters from the environmental monitoring module, and generates a dynamic fishing strategy. The fishing strategy includes the adjustment parameters of the fishing net mesh size, the coordinates of the deployment position, the fishing duration, and the timing of retrieval. The fishing execution module includes an adjustable mesh intelligent fishing net, an underwater positioning unit, a drive unit, and a retrieval unit. The adjustable mesh intelligent fishing net achieves stepless adjustment of the mesh size from 5 to 20 cm by extending and retracting the mesh wires via a motor. The wireless communication module uses a combination of 4G / 5G and underwater acoustic communication to realize data transmission between the central control module and remote terminals and various functional modules, with a transmission delay of ≤1s; The power module provides stable power to each module, including a 12V / 100Ah waterproof lithium battery pack and a 100W solar charging unit. It has overcharge, over-discharge, and short-circuit protection functions, and the battery life is ≥72 hours.

2. The intelligent fishing system based on image recognition according to claim 1, characterized in that, The improved YOLOv8 deep learning model adds CBAM (Convolutional Block Attention Module) to the backbone to enhance the extraction of key features of fish fins and scales; and uses BiFPN (Bidirectional Feature Pyramid Network) to optimize multi-scale feature fusion in the neck part.

3. The intelligent fishing system based on image recognition according to claim 1, characterized in that, The fish feature database contains standard images and annotation information for more than 30 species of freshwater and saltwater fish. The annotations include fish body length, body width, outline features, growth stage, and typical texture information, and support incremental training and updates based on fishing feedback data.

4. The intelligent fishing system based on image recognition according to claim 1, characterized in that, The underwater positioning unit uses a combination of GPS and underwater acoustic positioning beacons, with a positioning accuracy of ≤1m. The drive unit includes two thrusters with a maximum power of 500W and an attitude adjustment mechanism, which enables fine-tuning of the fishing net's attitude in the horizontal and vertical directions with an adjustment accuracy of ≤0.5°.

5. The intelligent fishing system based on image recognition according to claim 1, characterized in that, The adjustable mesh smart fishing net uses 304 stainless steel wire and a waterproof stepper motor drive mechanism, and the net's unfolded area can be 10-50m². 2 The mesh size can be adjusted as needed within the specified range, with an adjustment accuracy of ≤0.5cm. The mesh surface is coated with a polytetrafluoroethylene anti-corrosion coating, and the underwater service life is ≥3 years.

6. An intelligent fishing method based on image recognition, applied to the system described in any one of claims 1-5, characterized in that, Includes the following steps: S1: Data acquisition starts. The underwater image acquisition module acquires underwater image data at a preset frequency of 5fps, and the ultrasonic-assisted imaging unit acquires contour supplementary data simultaneously. The environmental monitoring module acquires water temperature, water quality, dissolved oxygen and water pressure parameters every 10s. All data are transmitted to the central control module in real time through the wireless communication module and stored in the local database. S2: Image preprocessing. After receiving the data, the image preprocessing module sequentially performs median filtering for noise reduction, improved dark channel prior dehazing, and CLAHE adaptive contrast enhancement. Finally, it generates an enhanced fish image through a weighted fusion algorithm. S3: Intelligent Recognition and Analysis. The intelligent recognition and analysis module calls the improved YOLOv8 model and combines it with a fish feature database to achieve species identification. It calculates body length / width using pixel calibration and statistically analyzes data from 10m... 2 The number and density of fish within the coverage area are predicted using the Kalman filter algorithm to determine the fish's movement trajectory in the next 5-10 seconds, with a positioning error of ≤0.5m. S4: Fishing strategy generation. The central control module combines the recognition results with environmental parameters to determine the fishing conditions. If the conditions are met, the net size, placement location and fishing duration are determined. If juvenile fish are present, the net size is increased by 5cm or the fishing area is adjusted. If the environmental parameters do not meet the standards, the fishing is delayed and an early warning is sent. S5: Automated fishing execution. The drive unit deploys the fishing net according to the positioning information and adjusts it to the preset mesh size. During the fishing process, the number of nets caught and the net's posture data are fed back in real time. The drive unit dynamically adjusts the net's posture according to changes in water flow. S6: Fishing and Recovery and Data Feedback. Recovery is initiated after the preset conditions are met. The actual fishing results are statistically analyzed and compared with the identification results. The identification accuracy is calculated, and the data is uploaded to the remote terminal and the fish feature database and model parameters are updated.

7. The intelligent fishing method based on image recognition according to claim 6, characterized in that, The pixel calibration method described in S3 is implemented as follows: a standard calibration plate is fixed in advance within the field of view of the underwater image acquisition module, and a nonlinear conversion model of pixel-actual size is established through calibration experiments at different water depths to eliminate underwater refraction errors.

8. The intelligent fishing method based on image recognition according to claim 6, characterized in that, The fishing conditions mentioned in S4 include: target fish body length ≥ preset commercial size, and aggregation density ≥ 2 fish / m². 3 The water temperature is 4-32℃, dissolved oxygen is ≥5mg / L, and pH is 6.5-8.

5. If any environmental parameter exceeds the threshold, the system will delay the harvesting for 1-2 hours and then retest. If the system fails to meet the standards after 3 tests, an emergency warning will be sent and the harvesting operation will be stopped.

9. The intelligent fishing method based on image recognition according to claim 6, characterized in that, The net posture adjustment logic in S5 is as follows: real-time water pressure data is collected by a water pressure sensor and the water flow velocity is calculated. When the water flow velocity is 0.3-0.5m / s, the posture adjustment mechanism tilts the net by 5-8°; when the water flow velocity is >0.5m / s, it tilts by 10-15° and increases the mesh size by 1-2cm.

10. The intelligent fishing method based on image recognition according to claim 6, characterized in that, The data feedback in S6 also includes: storing the feedback data on fishing efficiency, resource density changes, and mesh size adjustment accuracy in association with environmental parameters to form an aquaculture fishing data archive, which supports remote terminal export and analysis, and provides data support for subsequent aquaculture density adjustment and feeding strategy optimization.

Images