基于弱边缘特征融合的羊脸识别

Sheep face recognition based on weak edge feature fusion

针对羊面部某些边缘特征由于受到光照等环境因素影响产生的弱边缘从而影响羊个体身份识别的问题,提出一种弱边缘特征提取方法。首先建立数据集:数据采自内蒙古鄂托克旗绒山羊养殖基地,共采集38只羊的2015张图像;对数据进行图像旋转、水平平移等诸多图像增强操作后使数据量扩大至原来的8倍,共计16120张图像。然后训练弱边缘特征融合网络与主干特征提取网络,对两个网络提取到的特征进行融合分类。在自建羊脸数据集上进行特征信息提取识别实验的结果表明,弱边缘特征融合网络Edge_AlexNet、Edge_VGG16、Edge_ResNet50网络识别精确度达到92.45%、98.15%、94.67%,比未融合弱边缘特征的AlexNet、VGG16、ResNet50分类精确度提高了10.09、13.48、6.64个百分点,可见融合弱边缘特征后的网络具有较高的精度与鲁棒性。

To address the problem that some edge features of sheep faces are affected by environmental factors such as lighting,thus affecting the identification of individual sheep,a weak edge feature extraction method was proposed Firstly,a dataset of 2015 images of 38 sheep was collected from a sheep farming base in Etoch Banner,Inner Mongolia,and secondly,the data volume was expanded by 8 times after many image enhancement operations such as image rotation and horizontal panning were performed on the data,totalling 16120 images.Finally,a weak edge feature fusion network and a backbone feature extraction network were trained to fuse the features extracted by the two networks for classification.The results of the feature information extraction and recognition experiments on the self-built sheep face dataset showed that the weak edge feature fusion networks Edge_AlexNet,Edge_VGG16,Edge_ResNet50 achieved recognition accuracies of 92.45%,98.15%and 94.67%,which were 10.09,13.48 and 6.64 percentage points higher than those of AlexNet,VGG16 and ResNet50 without fusion of weak edge features.It is shown that the networks with fusion of weak edge features have higher accuracy and robustness.