YOLO object detection method for underwater trash collecting robot: Octacleaner

Abstract

Million tons of trash is generated every day, which is often inadequately managed and sent to landfills. When it rains, trash along with agricultural waste and sediment, washes into water bodies, leading to underwater pollution. The heavy plastics, metals, glass, and organic waste settle at the water body’s bottom, compounding the problem. To solve this, we’ve developed the Octacleaner, an underwater trash-collecting robot equipped with a YOLO object detection model. In our study, we compared the performance of YOLOv5 and YOLOv8. YOLOv5 boasts efficient and rapid object detection, making it suitable for real-time applications. It utilizes a streamlined architecture, making it computationally efficient while maintaining accuracy. YOLOv8, on the other hand, takes this a step further, achieving high accuracy with mAP of 0.84, offering faster and highly accurate results. Its strength lies in its ability to swiftly and effectively identify objects. However, the unique challenge in underwater environments is that trash can change its shape due to water currents and decomposition. Additionally, factors like reduced sunlight and the presence of underwater mud can make predictions more challenging. Despite this, YOLOv8 shows promise in addressing these difficulties, with its superior accuracy and speed.