基于图像特征的模板匹配成鱼多目标识别算法研究与程序设计  

Research and program design of multi-target recognition algorithm for fish based on image feature template matching

作　　者：冯书庆 王志勇[1] 谌志新[1] 徐志强[1] FENG Shuqing;WANG Zhiyong;CHEN Zhixin;XU Zhiqiang(Fishery Machinery and Instrument Research Institute,Chinese Academy of Fishery Science,Shanghai 200092,China)

机构地区：[1]中国水产科学研究院渔业机械仪器研究所,上海200092

出　　处：《渔业现代化》2023年第2期74-84,共11页Fishery Modernization

基　　金：国家重点研发计划课题“远洋渔船节能技术及捕捞装备自动化控制系统集成示范(2020YFD0901201)”。

摘　　要：为解决渔业养殖及转运自动计数要求,提出一种基于机器视觉的鱼类识别算法设计与程序编制,以证明使用数字图像识别分类技术的尺度不变特征变换SIFT及加速稳健特征SURF算法等可有效地检测与标注鱼类图像特征点。采用快速近邻匹配FLANN匹配算法测试了基于图像特征的鱼类旋转和泛化目标检测,得出SURF特征对个体检测效果好、SIFT特征对泛化目标检测效果优的结果。考察FLANN图像匹配碎片化特性、结合图像信息区域聚集实际、借鉴模板检测方法,设计了图像分割扫描及特征匹配的模板检测算法,并使用最大稳定极值区域MSER方法对识别结果进行冗余排除,达到能正确识别多目标鱼类的算法预设目标。研究发现,该算法及软件能成功识别图片中的多个鱼类目标,测试效果好,有较强的实用意义。To solve the need for automatic counting of fishery breeding and transportation,a fish identification algorithm design and programming based on machine vision is proposed.Using digital image recognition and classification technology,proved that Scale Invariant Feature Transform SIFT and Speed Up Robust Feature SURF algorithm can effectively detect and label fish image feature points.A Fast Library for Approximate Nearest Neighbors FLANN matching algorithm is designed to test rotation and generalized fish target finding based on image features proving that the SURF feature is good for individual detection and SIFT feature is good for generalization target detection.In view of the fragmentation characteristics of FLANN image feature matching,combined with the actual situation of image information area aggregation,and using the template detection method for reference,a template detection algorithm for image segmentation scanning and feature matching is designed;using Maximally Stable Extremal Regions MSER method to eliminate the redundancy of the recognition results,test result proved that the algorithm can correctly identify the multi-fish target.This study found that the algorithm and software could successfully identify multiple fish targets in the images,with good test results and strong practical implications.

关 键 词：目标识别 图像SIFT特征 图像SURF特征 FLANN特征匹配 特征模板匹配 鱼类 

分 类 号：S951.2[农业科学—水产养殖]