YOLOv8-RepGhostEMA: An efficient underwater trash detection model

Abstract

Abstract The accumulation of anthropogenic waste in underwater environments leads to a decrease in water quality, resulting in pollution that negatively impacts human health, ecological balance, and economic activities. The development of underwater robotic technology signals a new era for the timely identification and removal of underwater trash, offering proactive measures to combat the menace of water pollution. This study introduces an improved algorithm based on YOLOv8, proposing the YOLOv8-RepGhost-EMA algorithm. This algorithm is tailored for lightweight modifications on embedded devices of underwater robots and innovatively incorporates the RepGhost-EMA module, enhancing the precision of underwater trash detection while reducing computational complexity. Experimental results demonstrate that the proposed method achieves a reduction in parameter quantity by 13% while increasing precision (P) by 4%, recall (R) by 1.1%, and mean average precision (mAP) by 1.9% for trash detection. This approach is hardware-friendly for underwater robots, achieving a lightweight underwater object detection algorithm that balances detection accuracy and speed.