Adaptation of YOLOv8n for Underwater Object Detection System

• Main Authors: Abu, Zurah, Zarawi, Siti Nur Aisyah Zulaikha Mohd, Aminuddin, Raihah, Sabri, Nurbaity, Hamzah, Raseeda, Sheng, Chew Chiou
• Format: Conference Proceeding
• Language: English
• Subjects: Adaptation models; Deep learning; image processing; Pollution; Process control; Real-time systems; Reliability; Streaming media; underwater object; Usability; Webcams; YOLO
• ISSN: 2769-7916
• DOI: 10.1109/ICSPC63060.2024.10862573

Underwater pollution is a serious issue that can affect our environment. This pollution often involves the contamination of harmful substances, such as garbage, which can impact marine ecosystems. Therefore, many initiatives have been introduced to reduce underwater pollution, such as manually collecting garbage. However, employing individuals for this task is expensive and poses health and safety risks to divers. To address this issue, technology using image recognition systems has been developed. Image processing can be integrated with deep learning technology to enhance the efficiency and effectiveness of these efforts. Therefore, a system for detecting underwater objects with deep learning model YOLO version 8 is developed to tackle these issues. A system utilizing the deep learning model YOLO version 8 has been developed to detect underwater objects. The system operates by capturing real-time video and images using a webcam, categorizing underwater objects based on their movement. The dataset used comprises images of goldfish, tortoises, and coins sourced from image.cv, Google Images, and Kaggle. A metric evaluation of the object detection model was conducted. The model achieved {9 9. 4 \%} mean average precision accuracy. This system is designed to aid governance bodies from both the public and private sectors in their environmental preservation efforts.