Marine Mine Detection Using Deep Learning

Abstract

The paper addresses the detection of floating and underwater marine mines from images recorded from cameras (taken from drones, submarines, ships, boats). Due to the lack of image datasets, images were taken from the Internet and by using the technique of augmentation and synthetic image generation (by overlapping images with different types of mines over water backgrounds) 2 data sets were built (one for floating mines and one for underwater mines). The networks were trained and compared using 3 types of Deep Learning models Yolov5, SSD and EfficientDet (Yolov5, SSD for floating mines and Yolov5 and EfficientDet for underwater mines). The networks were also tested in the context of an IoT device (RaspberryPi 4, RPi camera).