Publication

Abstract : Floating trash like bottles are one of the main causes of the serious environmental threat posed by the growing pollution of inland water bodies. Conservation efforts are hampered by the lack of effective and affordable techniques for locating and retrieving these bottles. Real-time object detection algorithms in conjunction with unmanned boats can offer ways to remove floating trash from rivers. A popular real-time object detection model is the You Only Look Once (Yolo) Series. Based on efficacy and accuracy rate, each YOLO model gets better than the other. The study presents an extensive analysis of YOLO designs, such as YOLOv5, YOLOv7, YOLOv8, and YOLO-NAS, for bottle identification in inland water. The dataset used is a floating trash in the river, and the performance metrics such as precision, recall and mean average precision are employed, to compare and experimentally validate each algorithm’s detection efficiency and accuracy. The methodology trains each architecture for 200 epochs. Regarding the precision of detection, the experimental results shows that YOLOv8-x model has high value of 85.5% and mAP@[IoU=0.5:0.95] of 43.5%. YOLO-NAS variations have high recall, they are applicable for minimizing missed detection. YOLO-NAS m has a higher recall value of 90%.