This paper aims to evaluate the state-of-the-art object detection network; YOLOv5s (You Only Look Once version 5 small) for the detection of underwater marine debris using AUVs. The development of machine learning and AUVs for detecting marine debris is reviewed. In the paper, the YOLOv5s model is trained on a marine debris dataset using transfer learning. Several other object detection models are also trained on the same dataset for comparison. The results of the trained models are evaluated and the YOLOv5s model is deployed on an Android device to determine its suitability for real-time marine debris detection onboard AUVs. Overall, the YOLOv5s was able to achieve high accuracy scores of up to 91.2% and fast detection speeds of up to 20FPS on a Poco X3 Pro. |
*** Title, author list and abstract as seen in the Camera-Ready version of the paper that was provided to Conference Committee. Small changes that may have occurred during processing by Springer may not appear in this window.