摘要
为了实现对猪的精准活体身份识别,基于现有的双线性卷积神经网络(Bilinear-CNN),提出了一种非侵入式的猪面部识别模型。利用在图像特征提取上具有优良效果的VGG-16网络作为特征提取器,并将不同层次的提取特征做外积融合以形成最终的个体身份特征,最后,利用全连接层对其进行分类。实验结果表明:识别模型能对不同光照、角度、表情和姿态的猪脸进行识别,在200头猪的2 110张测试图像集中,识别准确率达到95.73%。
In order to realize live identification accurately of pigs, a non-invasive facial recognition model was proposed based on the existing Bilinear-CNN. The model uses VGG-16 networks with excellent effects on image feature extraction as feature extractors, and combines the different levels of extracted features into outer product to form the final individual identity feature. Then classify it using a fully connected layer. Experimental results show that the recognition model can identify pig face with different lighting, angles, expressions and postures with 95.73% accuracy on the test sets containing 2 110 images of 200 pigs.
作者
秦兴
宋各方
QIN Xing;SONG Gefang(School of Electronic Information,Hangzhou Dianzi University,Hangzhou Zhejiang 310018,China)
出处
《杭州电子科技大学学报(自然科学版)》
2019年第2期12-17,共6页
Journal of Hangzhou Dianzi University:Natural Sciences
基金
科技部云计算与大数据资助项目(2016YFB1000400)
关键词
猪脸识别
细粒度分类
卷积神经网络
多层次融合
pig face recognition
fine-grained classification
convolutional neural network
multi-level integration