基于YOLOX的声呐图像目标检测  

Sonar Image Target Detection Based on YOLOX

作　　者：程娟[1] 田梅[1] 柳晶晶 张浩庭 董春宵[1] CHENG Juan;TIAN Mei;LIU Jingjing;ZHANG Haoting;DONG Chunxiao(Information Engineering University,Zhengzhou 450001,China)

机构地区：[1]信息工程大学,河南郑州450001

出　　处：《信息工程大学学报》2023年第4期385-390,共6页Journal of Information Engineering University

基　　金：国家自然科学基金资助项目(62072057)。

摘　　要：近年来,随着人工智能技术的发展,深度学习越来越多地应用于声呐图像目标检测领域,逐渐成为近海国防建设和海洋经济发展中的热点研究课题。针对声呐图像分辨率低、边缘细节模糊、斑点噪声严重等问题,提出了一种基于YOLOX模型的声呐图像目标检测算法。同时,考虑到声呐图像获取困难、数据集通常较小,又引入迁移学习的方法对预训练模型进行微调,得到适用于声呐图像的目标检测模型。最后,在URPC2021数据集上进行实验,以验证该算法的检测性能。In recent years,with the development of artificial intelligence,the technology of deep learning has been applied in sonar image target detection and become a hot topic gradually in marine military defense and economic development.To address such problems as low resolution,blurred edge and speckle noise of sonar images,a sonar image target detection algorithm based on YOLOX is proposed.Then,a migration learning method is introduced to solve the problem of small image dataset.Finally,experiments are performed on URPC2021 dataset to verify the performance of the proposed algorithm.

关 键 词：目标检测 声呐图像 深度学习 迁移学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]