Underwater target feature extraction and posture estimation based on deep learning

With the development of the deep learning theory, the rapid and accurate target recognition can be achieved through the training of deep convolutional neural networks. In this paper, a deep learning-based underwater target feature extraction and posture estimation approach for estimating the posture information of the target using only visual images. The proposed approach improves the network structure of the YOLOV3 algorithm for the feature extraction of the target images and introduces characteristic marks on the underwater target for the posture estimation. Experiment results on a common tool, i.e., wrench under the water show that the mAP of the target detection is as high as 99.41% and the IOU is 87.03%, which show the performance of the proposed underwater target feature extraction and posture estimation approach.