水下图像目标检测技术研究  

Research on Underwater Image Target Detection Technology

作　　者：方俊[1] 邱春荣[1] FANG Jun;QIU Chunrong(Software School of Changsha Social Work College,Changsha,Hunan Province,410004 China)

机构地区：[1]长沙民政职业技术学院软件学院,湖南长沙410004

出　　处：《科技资讯》2022年第12期26-29,共4页Science & Technology Information

基　　金：湖南省教育厅科学研究课题(项目编号:19C0100)。

摘　　要：运用目标检测技术,水下机器人海底打捞技术可替代人工捕捞,解决人工作业危险系数高、经济效益低的问题。该文水下目标检测研究对象为海参、海胆、扇贝、海星四类生物,针对水下图像颜色偏移严重、存在明显跨域等问题,提出了一种基于Cascade-rcnn的海底生物目标检测方法,结合传统图像处理,解决水下图像跨域问题,最终精度达到0.507(iou0.5:0.95 Map)。Using target detection technology, underwater robot seabed fishing technology can replace manual fishing and solve the problems of high risk coefficient and low economic benefit of manual operation.The research objects of underwater target detection in this paper are four types of organisms: sea cucumbers, sea urchins, scallops,and starfish. Aiming at the problems of serious color shift and obvious cross domain in underwater images, this paper proposes a seabed biological target detection method based on Cascade-rcnn. Combined with traditional image processing, the cross domain problem of underwater images is solved, and the final accuracy reaches 0.507(iou0.5:0.95 Map).

关 键 词：目标检测 Cascade-rcnn 跨域 颜色偏移 

分 类 号：TP29[自动化与计算机技术—检测技术与自动化装置]