A single-loop eel farming photo-based resource management system based on cmyolov8

Abstract

This invention discloses a CMYOLOv8-based image resource management system for *Ulva pertusa* aquaculture, belonging to the field of digital technology for artificial breeding of *Ulva pertusa*. The system adopts a B / S architecture, with the backend built on the Flask framework for image processing, embedding a CMYOLOv8 model to achieve intelligent detection and processing of aquaculture images. The frontend uses HTML, CSS, and JavaScript to develop interactive pages, supporting user login, batch image upload, image classification, and visualization of burrow number change curves. The system uses an SQLite database to store aquaculture images and burrow monitoring data, relying on a RESTful interface based on the HTTP protocol to complete standardized data interaction between the front and back ends. Image data is processed by calling the model through the webpage, and the results are returned to the frontend for rendering and display in JSON format. This invention realizes intelligent processing and visual management of *Ulva pertusa* aquaculture image resources, making up for the shortcomings of traditional digital management of aquaculture, and providing a convenient and efficient technical means for aquaculture monitoring and data statistical analysis.

Description

Technical Field

[0001] This invention relates to the field of digital artificial breeding of Ulva unicinctus, and more specifically, to a resource management system for Ulva unicinctus breeding photos based on CMYOLOv8. Background Technology

[0002] The artificial aquaculture industry of *Ulva unicinctus*, commonly known as sea cucumber, has been expanding year by year due to continuously increasing market demand, becoming an important pillar industry for local aquaculture income and fishery economic development. Under the large-scale aquaculture model, the statistics of sea cucumber burrow numbers, monitoring of aquaculture status, and traceability of growth data are the core keys to aquaculture production management, yield prediction, and disease control, directly determining the level of refined aquaculture management and industrial economic benefits.

[0003] Currently, the management of *Ulva unicinctus* aquaculture in China still relies primarily on traditional manual methods, which suffer from numerous technical and management shortcomings and are ill-suited to the demands of large-scale, intelligent aquaculture development. Firstly, counting sea cucumber burrows depends entirely on manual visual statistics, which is not only inefficient and costly, but also highly subjective and susceptible to interference from factors such as beach lighting, terrain, and high-density aquaculture environments, leading to significant statistical errors and failing to provide accurate data support for yield calculations and population density control. Secondly, the on-site image data collected during aquaculture is mostly stored in a scattered manner, lacking standardized classification mechanisms, rapid retrieval functions, and a full-process traceability management system. This results in disorganized and unreliable data, low data utilization, and difficulty in achieving digital control and long-term data analysis of the aquaculture process.

[0004] In terms of intelligent detection technology applications, existing aquatic detection solutions mostly use the traditional YOLOv8 model, which has poor adaptability when applied to complex aquaculture scenarios in tidal flats. On the one hand, the chaotic background of tidal flats, the dense distribution of sea cucumber burrows with high feature similarity, easily leads to missed detections and false detections in the model, and the detection accuracy cannot meet production needs. On the other hand, the model has a large number of parameters, redundant structure, slow inference speed, and high hardware deployment threshold, making it difficult to adapt to the lightweight and real-time deployment requirements of aquaculture resource management systems, and thus unable to be implemented in actual intelligent aquaculture operations.

[0005] Furthermore, current research on intelligent identification in aquatic products, both domestically and internationally, largely focuses on individual identification and detection of common aquatic products such as sea cucumbers and scallops. There is a lack of specialized research on the detection of burrows in *Ulva pertusa*. Existing technical solutions are generally limited in function, only capable of basic identification and detection, and have not yet formed a comprehensive integrated management system encompassing "intelligent identification, data storage, accurate statistics, and trend analysis." Simultaneously, existing models generally suffer from a technical contradiction between accuracy and lightweight design, lacking sufficient adaptability and practicality for various scenarios. This fails to effectively address the pain points of intelligent management in the large-scale farming of *Ulva pertusa*, severely hindering the digital and intelligent transformation and upgrading of the sea cucumber aquaculture industry. Summary of the Invention

[0006] To address the aforementioned problems, the present invention aims to provide a resource management system for Ulva monocranium aquaculture photos based on CMYOLOv8, which addresses the issues of low efficiency in manual counting of Ulva monocranium burrows, chaotic image management, poor model adaptability, and limited functionality in existing technologies.

[0007] To achieve the above technical objectives, this application provides a resource management system for *Ulva unicornu* aquaculture photos based on CMYOLOv8, comprising: The batch upload and classification module is used to upload multiple photos of *Ulva unicornu* farming, classify them, and save them to the database. The intelligent hole-entry recognition module is used to identify and count hole entrances based on the CMYOLOv8 model by calling up photos of *Ulva unicornu* farming from the database, generating labeled images, and storing them in the database. The history management module is used to retrieve, view, and delete historical images and recognition results from the database; The trend analysis module is used to generate a line chart of changes in the number of openings based on historical data in the database.

[0008] Preferably, the batch upload and classification module is used to classify and store data according to aquaculture scenario, collection time, and water quality indicators, wherein the water quality indicators include one or more combinations of temperature, pH, and dissolved oxygen.

[0009] Preferably, the database is an SQLite database.

[0010] Preferably, the database is MySQL / PostgreSQL.

[0011] Preferably, the CMYOLOv8 model is based on the YOLOv8 framework, with the backbone network using the CSPHet module combined with heterogeneous kernel convolution, and the neck network using the MetaNeXt structure combined with Inception multi-branch.

[0012] Preferably, the intelligent hole recognition module is also used to construct a dataset adapted to real aquaculture scenarios for model training. The dataset consists of hole images with different lighting, mud and sand textures and distribution densities, which are uniformly processed to 640×640 pixels and labeled using data augmentation.

[0013] Preferably, the CMYOLOv8 model is trained using the PyTorch framework, with 100 iterations, a batch size of 8, and an initial learning rate of 0.01. Cosine annealing and early stopping mechanisms are used to prevent overfitting.

[0014] Preferably, the optimal confidence threshold for the CMYOLOv8 model is 0.25.

[0015] Preferably, the system also includes a user login module for verifying data security through account and password.

[0016] Preferably, the system also includes a data export module for exporting the recognition results in CSV format.

[0017] The present invention discloses the following technical effects: This invention integrates functions such as automatic hole identification and counting, image standardization management, data statistical analysis, trend visualization, and data export, thereby improving aquaculture management efficiency, adapting to complex tidal flat / pond aquaculture scenarios, and promoting the digital transformation of aquaculture. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of the system architecture described in this invention.

[0020] Figure 2 This is a comparison chart of the hole identification effect in laboratory and aquaculture environments as described in this invention.

[0021] Figure 3 This is a schematic diagram of the marking operation of the sea cucumber cavities described in this invention.

[0022] Figure 4 This is the curve showing the change in loss and performance indicators during the model training process described in this invention.

[0023] Figure 5 This is a visualization of the detection effect described in this invention.

[0024] Figure 6 This is the ER diagram of the database structure described in this invention.

[0025] Figure 7 This is a schematic diagram of the system execution flow described in this invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0027] like Figure 1 As shown, this invention provides a resource management system for CMYOLOv8-based aquaculture photos of *Ulva unicornu*. The system adopts a B / S architecture, with the backend based on the Flask framework and SQLite database, integrating a lightweight CMYOLOv8 model; the frontend uses HTML / CSS / JavaScript to develop the interactive interface.

[0028] For example, the B / S architecture, or Browser / Server architecture, is currently the mainstream web application architecture pattern.

[0029] For example, the Flask framework is a lightweight web framework written in Python.

[0030] For example, SQLite database is an ultra-lightweight, installation-free embedded database.

[0031] For example, the front end can also be developed into a mini-program, supporting real-time uploading and recognition on mobile devices.

[0032] For example, the backend can also use the Django framework.

[0033] In one embodiment, the system consists of a user login module, a batch upload and classification module, an intelligent hole recognition module, a historical record management module, a trend analysis module, and a data export module.

[0034] In one implementation, the user login module is used for account and password verification to ensure data security.

[0035] In one implementation, the batch upload and classification module is used to perform multi-image uploads and classify and store them according to aquaculture scenario (indoor pond / outdoor tidal flat), collection time, and water quality indicators (temperature / pH / dissolved oxygen).

[0036] In one embodiment, the intelligent hole recognition module is used to call the CMYOLOv8 model to automatically identify and count holes, generate labeled images, and display the results in real time.

[0037] In one implementation, the CMYOLOv8 model is based on YOLOv8 and features lightweight improvements to address issues such as small targets at the entrance of a single-ringed eel's burrow, complex backgrounds, and uneven illumination. The backbone network adopts a CSPhet structure and introduces heterogeneous kernel convolutions to effectively reduce the number of parameters and computational cost; the neck network is replaced with a MetaNeXt structure and combined with an Inception branch to enhance the feature extraction capability for small targets.

[0038] For example, the CMYOLOv8 model is based on the YOLOv8 framework and has been specifically optimized. It is divided into four parts: input layer, backbone feature extraction network, neck feature fusion network, and detection head. The backbone network abandons the original structure and adopts the CSPHet module with heterogeneous kernel convolution to achieve model lightweighting while ensuring basic feature extraction capabilities, effectively reducing the number of parameters and computational overhead. The neck network uses the MetaNeXt structure combined with Inception multi-branch to take advantage of the multi-receptor field feature extraction to solve the detection problems caused by complex backgrounds, uneven lighting, and small target size of sea cucumber holes in tidal flat environments.

[0039] For example, after the model front end completes image preprocessing, the backbone network completes preliminary feature extraction, the neck network then fuses and enhances multi-scale features, and finally the YOLOv8 detection head completes target localization and classification, outputting the detection results and statistical count of single-ringed eel burrows.

[0040] For example, such as Figure 2 As shown in the preliminary experiments, the original model performs well only in the laboratory environment and has poor recognition performance in actual aquaculture scenarios, requiring further training and optimization.

[0041] For example, such as Figure 3 As shown in Table 1, this invention constructs a dataset adapted to real aquaculture scenarios, collecting 176 images of burrows with different lighting, mud and sand textures, and distribution densities. Data augmentation was performed through flipping, brightness adjustment, and other methods. The images were uniformly processed to 640×640 pixels and labeled, resulting in a total of 3978 burrow instances labeled. The dataset was divided into training and validation sets in an 8:2 ratio to provide sufficient and diverse sample support for model training.

[0042] Table 1 training set 141 80% Model learning Validation set val 35 20% Hyperparameter modification and verification total 176 For example, the model designed in this invention is trained using the PyTorch framework, with 100 iterations, a batch size of 8, and an initial learning rate of 0.01. Cosine annealing and early stopping mechanisms are used to prevent overfitting. Hyperparameters are optimized through multiple sets of comparative experiments, and 0.25 is finally selected as the optimal confidence threshold. After training, the model has a Precision of 0.480, a Recall of 0.552, and an mAP50 of 0.486 on the validation set. The loss curve converges smoothly, and the single-graph inference time is 233.9ms, which can meet the needs of rapid hole identification in actual aquaculture scenarios.

[0043] In one implementation, after model training and hyperparameter modification, the final stage uses an independent test set to comprehensively validate the final CMYOLOv8 model. Evaluation metrics include mean accuracy (mAP50, mAP95), number of parameters, computational speed, false negative rate, and false positive rate. The experimental results are obtained through comparative and ablation experiments.

[0044] For example, through analysis of relevant metrics, the box_loss and cls_loss decreased and stabilized after approximately 60 training rounds, indicating that the model has initially acquired a basic ability to identify the opening of a single-ringed eel, meeting the most basic identification requirements. Figure 4 , Figure 5 As shown in Table 2.

[0045] Table 2 Summary of evaluation index values ​​corresponding to training results numerical values 100 2.321 1.637 0.48 0.552 0.486 0.2 233.9ms Optionally, to adapt to different computing power terminals, the model can use lightweight models such as YOLOv8n / YOLOv10s to identify and count holes.

[0046] In one implementation, the history management module is used to retrieve / view / delete historical images and recognition results in the database, supporting traceability.

[0047] For example, such as Figure 6 As shown, the database consists of a user table (USER), an image record table (IMAGE), and a recognition result table (RESULT). The user table (USER) stores account information, passwords, and access control. The image record table (IMAGE) stores file names, paths, upload times, classifications, and water quality indicators. The recognition result table (RESULT) stores the number of openings, recognition time, and associated image IDs. As shown in the figure, the three tables are linked to ensure data consistency.

[0048] Optionally, to accommodate large-scale deployments, MySQL / PostgreSQL may be chosen as the database.

[0049] In one implementation, the trend analysis module is used to generate a line graph of changes in the number of burrows based on historical data, which visually displays the dynamics of aquaculture.

[0050] In one implementation, the data export module is used to support the export of recognition results in CSV format for easy subsequent analysis.

[0051] In summary, the system designed in this invention adopts a B / S (Browser / Server) architecture for overall design. The backend is an image processing system based on the Flask framework, embedding the CMYOLOv8 model within it, and also controlling data input and output. The frontend uses a browser to visualize the backend data, employing HTML, CSS, and JavaScript technologies for page design. The visualization results include a user login interface, a batch image upload interface, a categorization interface, and a curve displaying the number of holes. The database uses an SQLite database to store image and hole quantity data. For model integration, a RESTful interface is used. A RESTful interface is a standardized front-end and back-end data interaction interface based on the HTTP protocol (a common rule for data transmission between browsers and servers), used to enable web pages to call the CMYOLOv8 model. After data upload and processing, the photos are sent to the frontend in JSON format for display. Figure 7 As shown, the system operates in the following mode: "batch image import, automatic classification and management, model recognition and counting, quantity change statistics, trend curve drawing, and data export". In addition, the system also supports subsequent data retrieval, export and analysis, which meets the needs of long-term monitoring and historical data traceability in the breeding process.

[0052] Compared with existing technologies, this invention automates the identification and counting of holes, replacing manual labor and improving efficiency by more than 80%. Single image recognition takes only 233.9ms and supports batch processing.

[0053] Compared with existing technologies, this invention adapts to complex tidal flat / pond environments using the CMYOLOv8 model, achieving an mAP50 of 0.486, effectively reducing the rate of missed / false detections and meeting the needs of aquaculture monitoring.

[0054] Compared with existing technologies, this invention enables the classification, storage, traceability retrieval, and data export of aquaculture images, solving the problems of chaotic management and data loss in traditional methods.

[0055] Compared with existing technologies, this invention automatically generates a trend chart of the number of burrows, which intuitively reflects changes in aquaculture density and assists in making refined aquaculture decisions.

[0056] Compared with existing technologies, the model of this invention has a small number of parameters and fast inference. The system is based on a B / S architecture, requires no client installation, and can be used with ordinary computers / browsers. It is low-cost and easy to promote.

[0057] Compared with existing technologies, this invention supports model replacement and functional module expansion, and can be adapted to other aquaculture scenarios such as sea cucumbers and scallops, making it highly versatile.

[0058] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0059] In the description of this invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0060] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A resource management system for CMYOLOv8-based aquaculture photos of *Ulva unicornu*, characterized in that, include: The batch upload and classification module is used to upload multiple photos of *Ulva unicornu* farming, classify them, and save them to the database. The intelligent hole-entry recognition module is used to identify and count hole entrances based on the CMYOLOv8 model by calling the photos of *Ulva unicornu* farming from the database, generating labeled images, and storing them in the database. The historical record management module is used to retrieve, view, and delete historical images and recognition results from the database. The trend analysis module is used to generate a line graph showing the change in the number of openings based on historical data in the database.

2. The resource management system for *Ulva unicornu* culture photos based on CMYOLOv8 as described in claim 1, characterized in that: The batch upload and classification module is used to classify and store data according to aquaculture scenario, collection time, and water quality indicators, wherein the water quality indicators include one or more combinations of temperature, pH, and dissolved oxygen.

3. The resource management system for *Ulva unicornu* culture photos based on CMYOLOv8 as described in claim 1, characterized in that: The database is an SQLite database.

4. The resource management system for *Ulva unicornu* culture photos based on CMYOLOv8 as described in claim 1, characterized in that: The database is MySQL / PostgreSQL.

5. The resource management system for *Ulva unicornu* culture photos based on CMYOLOv8 as described in claim 1, characterized in that: The CMYOLOv8 model is based on the YOLOv8 framework. The backbone network uses the CSPHet module with heterogeneous kernel convolution, and the neck network uses the MetaNeXt structure combined with Inception multi-branch.

6. The resource management system for *Ulva unicornu* culture photos based on CMYOLOv8 as described in claim 1, characterized in that: The intelligent hole recognition module is also used to construct a dataset adapted to real aquaculture scenarios for model training. The dataset consists of hole images with different lighting, mud and sand textures and distribution densities, which are uniformly processed to 640×640 pixels and labeled using data augmentation.

7. The resource management system for *Ulva unicornu* culture photos based on CMYOLOv8 as described in claim 1, characterized in that: The CMYOLOv8 model was trained using the PyTorch framework, with 100 iterations, a batch size of 8, and an initial learning rate of [missing information]. 0.01, cosine annealing and early stopping mechanism are used to prevent overfitting.

8. The resource management system for *Ulva unicornu* culture photos based on CMYOLOv8 as described in claim 7, characterized in that: The optimal confidence threshold for the CMYOLOv8 model is 0.

25.

9. The resource management system for *Ulva unicornu* culture photos based on CMYOLOv8 as described in claim 1, characterized in that: The system also includes a user login module for account and password verification to ensure data security.

10. The resource management system for *Ulva unicornu* culture photos based on CMYOLOv8 according to claim 1, characterized in that: The system also includes a data export module for exporting the recognition results as CSV format.

Images