基于深度学习的水下目标声学识别与定位技术研究  

Research on Underwater Target Recognition and Location Technique Based on Deep Learning

作　　者：岳成海 王旭[1,2,3] 宫俊玲[1,2,3] 曾俊宝[1,2,4] 徐高朋 YUE Chenghai;WANG Xu;GONG Junling;ZENG Junbao;XU Gaopeng(State Key Laboratory of Robotics,Shenyang Institute of Automation,Chinese Academy of Sciences,Shenyang 110016,China;Institutes for Robotics and Intelligent Manufacturing,Chinese Academy of Sciences Shenyang 110169,China;Key Laboratory of Opto-Electronic Information Technology,Chinese Academy of Sciences,Shenyang 110169,China;Key Laboratory of Marine Robotics,Liaoning Province,Shenyang 110169,China)

机构地区：[1]中国科学院沈阳自动化研究所机器人学国家重点实验室,辽宁沈阳110016 [2]中国科学院机器人与智能制造创新研究院,辽宁沈阳110169 [3]中国科学院光电信息处理重点实验室,辽宁沈阳110169 [4]辽宁省水下机器人重点实验室,辽宁沈阳110169

出　　处：《数字海洋与水下攻防》2021年第6期492-497,共6页Digital Ocean & Underwater Warfare

基　　金：国家自然科学基金面上项目(42176194);国家重点研发计划资助项目(2017YFC0821204)。

摘　　要：为实现自主水下潜器(Autonomous Underwater Vehicle,简称AUV)的自主目标探测识别与定位任务,以侧扫声呐数据为依据,考虑到扫描式声呐成像的特点,针对金属球类目标,基于Darknet框架设计了一种轻量化深度学习目标识别模型,并结合人工特征进行目标特性分析。同时对声呐图像设计了有效的图像增强方法。实验表明:上述目标识别方法在保证目标识别准确率的同时,具有较高的目标识别速率,适于低功耗嵌入式平台部署。In order to realize autonomous target detection,recognition and positioning task of autonomous underwater vehicle(AUV),basing on the side-scan sonar data and considering the characteristics of side-scan sonar images,a lightweight deep learning target recognition model is designed in this paper based on Darknet framework for metal balls.And the characteristics of metal balls are analyzed combining with artificial characteristics.At the same time,an effective image enhancement method is designed for sonar images.Experiments show that the objects recognition method described in this paper not only ensures the accuracy of target recognition,but also provides a high processing rate,which is suitable for the low-power embedded platform.

关 键 词：目标识别 声图合成 侧扫声呐 深度学习 Darknet框架 

分 类 号：TP242[自动化与计算机技术—检测技术与自动化装置]