Publication

Abstract :

Underwater naval mines are extremely hazardous to battleships and submarines. Therefore, naval forces around the world use mine countermeasures to protect their borders against such threats. One of the strategies employed in mine countermeasures is mine hunting, which requires a thorough search for all mines within a potentially hazardous region. This process is often segmented into four phases: detection, classification, identification, and disposal. In recent years, there has been growing interest in using underwater computer vision for inspection and monitoring purposes. In this paper, we conducted an experiment in a pool and collected a dataset of underwater naval mine-like target using an optical camera for classification. Here, we propose an Edge AI system, including an NVIDIA AGX Xavier device, for classifying custom underwater naval mine-like targets in images using the YOLOV5 deep learning object classification algorithm. We evaluated the model’s performance by analyzing the training loss, testing loss, and accuracy for 50 and 100 epochs. The results show that, for 100 epochs, the losses are minimal and the accuracy is >90%. Additionally, we analyzed the image processing time with the Edge AI device, averaging 216.7ms and 213.8ms for batches of 8 and 16 images.