基于小波变换和改进KPCA的奶牛个体识别研究
发布时间：2018-10-05 09:26
【摘要】：为加快畜牧业现代化程度,克服传统方法中奶牛个体识别正确率低的缺陷,针对奶牛个体纹理特征,对传统KPCA(核主成分分析)方法从降低协方差矩阵维数和引入类别信息两个角度进行改进,并与小波变换进行结合,应用于奶牛个体识别领域。首先对归一化后的奶牛图像进行一层小波分解得到4个分量子图,然后对各子图利用改进的KPCA进行特征提取并引入加权策略融合,最后构造出多类SVM分类器进行学习分类。将预先采集的20头奶牛个体的视频数据转化成图片序列并选取20 000张组成实验数据集,通过多组对比实验对小波融合系数、融合向量组数、特征维数三个重要参数进行设定,然后利用不同算法进行奶牛个体识别实验。结果表明,提出方法在识别正确率达到96.31%时,仅用了4.20 s,较其他算法具有明显优势,可以有效地应用到奶牛个体识别领域,兼具高性能、低成本的优势。
[Abstract]:In order to speed up the modernization of animal husbandry and overcome the defect of low correct rate of individual identification in traditional methods, this paper aims at the individual texture features of dairy cattle. The traditional KPCA (Kernel Principal component Analysis) method is improved from the aspects of reducing the dimension of covariance matrix and introducing category information. It is combined with wavelet transform and applied to the field of individual identification of dairy cows. Firstly, four sub-quantum graphs are obtained by wavelet decomposition of the normalized dairy cow image. Then, the improved KPCA is used to extract the features and the weighted strategy is introduced to each sub-graph. Finally, a multi-class SVM classifier is constructed for learning classification. The video data of 20 cows were converted into image sequence and 20 000 pieces of experimental data were selected. Three important parameters of wavelet fusion coefficient, fusion vector group number and feature dimension were set by comparison experiments. Then the dairy cow individual recognition experiment is carried out by different algorithms. The results show that when the recognition accuracy reaches 96.31, the proposed method only uses 4.20 s, which has obvious advantages over other algorithms, and can be effectively applied to the field of individual identification of dairy cows, with the advantages of high performance and low cost.
【作者单位】： 河北工业大学计算机科学与软件学院;河北省大数据计算重点实验室;
【基金】：天津市科委科技支撑计划项目(15ZCZDNC00130)
【分类号】：S823;TP391.41
本文编号：2252857
[Abstract]:In order to speed up the modernization of animal husbandry and overcome the defect of low correct rate of individual identification in traditional methods, this paper aims at the individual texture features of dairy cattle. The traditional KPCA (Kernel Principal component Analysis) method is improved from the aspects of reducing the dimension of covariance matrix and introducing category information. It is combined with wavelet transform and applied to the field of individual identification of dairy cows. Firstly, four sub-quantum graphs are obtained by wavelet decomposition of the normalized dairy cow image. Then, the improved KPCA is used to extract the features and the weighted strategy is introduced to each sub-graph. Finally, a multi-class SVM classifier is constructed for learning classification. The video data of 20 cows were converted into image sequence and 20 000 pieces of experimental data were selected. Three important parameters of wavelet fusion coefficient, fusion vector group number and feature dimension were set by comparison experiments. Then the dairy cow individual recognition experiment is carried out by different algorithms. The results show that when the recognition accuracy reaches 96.31, the proposed method only uses 4.20 s, which has obvious advantages over other algorithms, and can be effectively applied to the field of individual identification of dairy cows, with the advantages of high performance and low cost.
【作者单位】： 河北工业大学计算机科学与软件学院;河北省大数据计算重点实验室;
【基金】：天津市科委科技支撑计划项目(15ZCZDNC00130)
【分类号】：S823;TP391.41
【相似文献】
相关期刊论文 前2条
1 梁钰超;冯玉凡;汪斌;;RFID无线射频识别技术在犬个体识别中的应用研究[J];养犬;2010年04期
2 孙国凤;开发和牛的个体识别用基因标记[J];生物技术通报;1995年05期
，本文编号：2252857
本文链接：https://www.wllwen.com/yixuelunwen/dongwuyixue/2252857.html
