Real-Time Underwater Trash Object Detection Using Enhanced YOLOv8

The accumulation of trash in aquatic and marine environments has severe and significant consequences on marine ecosystems, resulting in a persistent environmental hazard. While volunteer trash pickup efforts have grown in recent years, they still demand considerable time, effort, and resources. To effectively address these challenges, the implementation of automated solutions is essential. This project aims to develop an efficient object detection deep learning model which helps in effectively identifying the underwater trash objects. The key challenges of underwater image-based localization come from object detection under low light conditions, also the presence of numerous small and overlapping targets, which significantly increase the complexity. In this project we have used improved Yolov8n object detection model achieving an impressive accuracy rate of mAP of 0.951. We have utilized diverse datasets covering a range of environmental conditions, including challenging low-light scenarios. Also, the integration of advanced preprocessing techniques and data augmentation has further enhanced our model’s performance, ensuring reliable predictions in various underwater environments, thereby improving detection performance without introducing additional computational complexity.