real-time localization of hard corals in underwater videos using darknet-yolo network

Real-time identifying and classifying hard corals on underwater videos is a critical task to cost-effectively monitor hard corals localization. in this work, we address the problem, using darknet-yolo framework. to this end, twenty-four convolutional layers of darknet-yolo is employed to detect a single hard coral class. the detection and localization method repeated on each video’s frame. to evaluated the framework performance, a collection of coral images has been extracted from the video and tagged manually. the collection consist 10000 sequential images and hard corals’ location are extracted on each frame. the system achieves approximately 88.3% on recall and 88.8% on accuracy.