Recognition of Fish in Aqua Cage by Machine Learning with Image Enhancement

Aim:

       The aim of this project is to propose a system to automate the process of fish population monitoring in aquaculture environments by utilizing the YOLOv8 deep learning-based object detection model, combined with image enhancement techniques. The proposed system is expected to improve the accuracy of fish detection and counting in underwater cages, offering an efficient, real-time solution for sustainable fish farming and ecological conservation.

Abstract:

     With the growing demand for fishery production and the depletion of capture fisheries resources, aquaculture has become an essential method for sustainable fish farming. Accurately monitoring fish populations in underwater cages is crucial, but traditional methods are labor-intensive and prone to errors. This study proposes a novel fish counting system using YOLOv8, a deep learning-based object detection model, to automate the process. By utilizing image enhancement and tracking algorithms, the system is expected to achieve an accuracy of up to 93%. The proposed system aims to provide an efficient, real-time solution for fishery resource management and ecological conservation.

Existing System:

      Current fish counting systems in aquaculture mostly rely on manual processes or basic machine learning techniques, both of which are limited by the unique challenges of underwater environments. Factors such as poor lighting, low visibility, and the erratic movement of fish hinder the accuracy of traditional methods. While more sophisticated algorithms like YOLOv4 have been applied to fish detection, they struggle with low generalization in these challenging underwater conditions. Image enhancement techniques such as Retinex have been explored to address some of these issues, but these methods are often complex, inflexible, and computationally expensive.

Problem Definition:

       Accurate fish counting in aquaculture is essential for effective resource management and ecological sustainability. Traditional methods of fish counting are labor-intensive and prone to human error. Furthermore, existing automated systems based on basic machine learning algorithms are hindered by the underwater environment’s complexities, such as poor visibility, lighting issues, and the behavior of fish. As a result, these systems fail to provide the level of precision required for reliable fish population monitoring in real-world conditions.

Proposed System:

      The proposed system aims to utilize the YOLOv8 deep learning model, a cutting-edge object detection technique, to accurately detect and count fish in real-time in underwater aquaculture environments. The system will consist of three main modules:

• Collect and prepare underwater fish cage images to create training, testing, and validation datasets, as well as the YAML file required for YOLO training.
• Train the YOLOv8 model on the collected dataset, with an expected best accuracy of up to 93%. The performance will be evaluated based on accuracy and loss rate.
• Implement a Flask-based web interface for real-time fish counting and tracking. The YOLOv8 model will process video frames in real-time, allowing operators to interact with the system and view live fish count and tracking data.

Advantage:

       The proposed system is expected to offer several advantages over traditional and existing automated methods. By targeting up to 93% accuracy, it aims to significantly improve fish detection and counting performance in underwater aquaculture environments. The use of YOLOv8 is anticipated to enable real-time processing, crucial for effective monitoring. The integration of Flask for a web-based interface will provide a user-friendly way for operators to interact with the system. Additionally, the system is designed to reduce human intervention, enhancing the efficiency of aquaculture management and contributing to sustainable fishery practices.