水下智能识别与自主抓取机器人设计与实现  

Design and implementation of underwater intelligent recognition and autonomous grasp vehicle

作　　者：高天铭 闫敬[1,2] 尤康林 张良 林景胜 罗小元 GAO Tian-ming;YAN Jing;YOU Kang-lin;ZHANG Liang;LIN Jing-sheng;LUO Xiao-yuan(Institute of Electrical Engineering,Yanshan University,Qinhuangdao Hebei 066004,China;Key Laboratory of Ocean Observation Technology,MNR,Tianjin 300110,China)

机构地区：[1]燕山大学电气工程学院,河北秦皇岛066004 [2]国家海洋技术中心自然资源部海洋观测技术重点实验室,天津300110

出　　处：《控制理论与应用》2022年第11期2074-2083,共10页Control Theory & Applications

基　　金：国家自然科学基金项目(62222314,61973263,61873345,62033011);自然资源部海洋观测技术重点实验室开放基金项目(2021klootA02);河北省青年拔尖人才项目(B J20200031);河北省杰出青年人才项目(F2022203001);河北省中央领导地方项目(226Z3201G);河北省三三三人才项目(C20221019)资助.

摘　　要：设计了一款面向海珍品捕捞的水下智能识别与自主抓取机器人.首先通过YOLOv4-tiny网络对海珍品图像离线训练,设计单双目自适应切换与多目标选择算法以实现海珍品在线识别与持续定位.进一步,采用声呐与深度传感器融合策略获取水下机器人深度信息,设计基于模糊比例–积分–微分控制的定深抓取控制器,以确保目标定位与抓取过程中深度信息的有效反馈.所提目标识别算法,具有实时性强、复杂度低优点;同时,定深与抓取控制器,不依赖于系统复杂模型,可适应不同海况下的精确抓取.最后,通过试验验证了方法的有效性.This paper presents an underwater intelligent recognition and autonomous grasp vehicle for the fishing of precious seafood.Firstly,we employ the YOLOv4-tiny network to off-line train the images of precious seafood,through which monocular-binocular adaptive switch and multi-target selection algorithms are conducted to achieve on-line recognition and persistent localization.Moreover,the depth information of the underwater vehicle is acquired by the fusion of sonar and depth sensor,and fuzzy proportional integral differential based depth-fixed and grasp controllers are designed respectively to guarantee effective feedback for the depth information during the procedures of target localization and grasp.The employed seafood recognition algorithm has the advantages of high real time and low complexity.On the other hand,the proposed depth-fixed and grasp controllers do not rely on complex system model,which can suit the target accurate grasp in different sea states.Finally,the experiments are conducted to verify the effectiveness of the developed approach.

关 键 词：识别 抓取 水下机器人 深度学习 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TP242[自动化与计算机技术—计算机科学与技术]