文献类型：会议论文

英文题名：Using Improved YOLOX for Underwater Object Recognition

作者：Zhang, Jun[1]; Peng, Xiaohong[1]; Zhang, Gaoyi[2]

机构：[1] Guangdong Ocean University, College of Mathematics and Computer Science, Zhanjiang, China; [2] University of Electronic Science and Technology, Postdoctoral Flow Station, Chengdu, China

会议论文集：2022 5th International Conference on Pattern Recognition and Artificial Intelligence, PRAI 2022

会议日期：August 19, 2022 - August 21, 2022

会议地点：Chengdu, China

语种：英文

外文关键词：Data handling - Image enhancement - Open Data

外文摘要：Underwater object recognition has limited sample data acquired and most of the images are of low resolution and poor quality resulting in less data that can be used for training. In this paper, we propose the underwater object recognition algorithm AX-YOLO based on YOLOX, which mainly improves the data enhancement mode in the data processing stage and the optimizer in network training according to the characteristics of under-water objects to test underwater object recognition in the case of open data sets underwater. The experimental results show that AX-YOLO model has the highest mAP of 91.05% and FPS of 71. The algorithm has a short training time and high accuracy, which supports underwater object recognition for marine operations development. ? 2022 IEEE.