YOLO-based Network Fusion for Riverine Floating Debris Monitoring System

Abstract

Riverine floating debris has been one of the major challenges and a well-known issue across the globe for decades. To mitigate this problem, sources of debris and their pathways to the riverine environment need to be identified and quantified. The scope of this study is to obtain visual information of floating debris which is crucial in developing a robotic platform for riverine management system. Therefore, an object detector using You Only Look Once version 4 (YOLOv4) algorithm is developed to detect floating debris for the riverine monitoring system. The debris detection system is trained on five object classes such as styrofoam, plastic bags, plastic bottle, aluminium can and plastic container. After the first training is conducted, image augmentation technique is implemented to increase training and validation datasets. Finally, the performance of the proposed debris detection system is evaluated based on the highest mean average precision (mAP) weight file, classification accuracy, precision and recall.