Position: Home > Articles > Research on recognition method of individual cattle muzzle based on local invariant features
Heilongjiang Animal Science and Veterinary Medicine
2022
(2)
48-52+136-137
基于局部不变特征的牛只个体唇纹识别方法研究
作 者：
王锋;李琦
单 位：
内蒙古科技大学信息工程学院
关键词：
牛唇纹;识别;预处理;关键点;SIFT特征;KNN特征匹配
摘 要：
为了改善传统方法牛只个体身份识别的误识别和操作复杂等现象,试验提出了一种基于计算机视觉提取牛唇纹图像局部不变特征来识别牛只的方法,即使用相机拍摄牛唇纹图像制作数据集,收集了51头牛的唇纹图像,每头有5～30张,共475张,对图像进行统一分辨率大小和限制对比度自适应直方图均衡化(contrast limited adaptive histogram equalization, CLAHE)预处理之后,使用关键点阈值为1 150个的尺度不变特征变换(scale invariant feature transform, SIFT)提取图像特征,最后用最近邻(K-nearest neighbors, KNN)特征匹配对牛唇纹图像分类,并比较了图像预处理、关键点数量阈值和SIFT、加速鲁棒特征(speeded-up robust features, SURF)、定向快速旋转(oriented fast and rotated brief, ORB)三种不同特征提取算法对牛唇纹识别准确率的影响。结果表明:使用统一分辨率大小和CLAHE处理牛唇纹图像,并设置关键点数量阈值为1 150个时,SIFT对牛唇纹图像识别效果最好,识别准确率为98.06%,平均反向负惩罚(mINP)为97.95%,置信度(confidence)为90.16%。说明采用计算机视觉识别个体牛身份是有效的,在牛唇纹的图像识别中,唇纹图像预处理、提取关键点的数量和特征提取算法对识别准确率有很大影响。
译 名：
Research on recognition method of individual cattle muzzle based on local invariant features
作 者：
WANG Feng;LI Qi;College of Information Engineering, Inner Mongolia University of Science and Technology;
关键词：
cattle muzzle;;recognition;;pretreatment;;keypoints;;SIFT features;;KNN feature matching
摘 要：
In order to improve the false identification and complex operation of individual cattle identification by traditional methods, a method based on computer vision to extract local invariant features of cattle muzzle pattern image was proposed in this paper.In the experiment, a camera was used to capture images of cattle muzzle to make a data set. A total of 475 muzzle cattle images of 51 cattle were collected, 5 to 30 for each.After preprocessing the image with uniform resolution size and contrast limited adaptive histogram equalization(CLAHE), the image features were extracted by scale invariant feature transform(SIFT) with key point threshold of 1 150. Finally, the cattle muzzle print image were classified by K-nearest neighbors(KNN)feature matching.The effects of image preprocessing, threshold of key points and SIFT,speeded-up robust features(SURF)and oriented fast and rotated brief(ORB) feature extraction algorithms on the accuracy of cattle muzzle print recognition were compared.The results showed that SIFT had the best recognition effect on cattle muzzle pattern image with 98.06% recognition accuracy, 97.95% mINP and 90.16% confidence when unified resolution size and CLAHE were used to process cattle muzzle pattern image and the number threshold of key points was set to 1 150. It indicated that computer vision was effective to recognize the identity of individual cattle. For the image recognition of cattle muzzle image, cattle muzzle image preprocessing, the number of key points and feature extraction algorithm had a great impact on the recognition accuracy.
相似文章
-
基于局部不变特征的牛只个体唇纹识别方法研究 [王锋, 李琦] 黑龙江畜牧兽医 2022 (1) 48-52,136-137
