基于Multi-Light模型的奶牛个体识别研究

Research on individual identification of cows based on Multi-Light model

为了解决大规模智能化奶牛养殖场对奶牛个体识别存在模型大、识别速度慢的问题,试验构建了一种用于识别奶牛个体的多尺度轻量化卷积神经网络(Multi-Light)模型,将拍摄的奶牛图像经过标注后利用DeepLab V3模型从复杂背景中分割出单头奶牛图像;在Multi-Light模型中引入空洞卷积,保证该模型参数量不变的同时增强提取图像全局信息的能力;加入多尺度卷积模块增强该模型对不同尺度特征点的检测能力,在该模型中使用短路连接以保证特征不丢失,提升模型的识别精度;此外,利用通道注意力机制提高了该模型识别精度,同时使该模型具有更多的非线性;最后将分割得到的奶牛图像数据集输入Multi-Light模型进行训练。结果表明:Multi-Light模型对奶牛个体识别的精度达98.51%,高于其他经典模型对奶牛个体的识别率;与轻量级模型对比,Multi-Light模型的大小为5.86 MB,在具备高识别精度的前提下参数量较少。说明试验所搭建的Multi-Light模型克服了传统方法中需要对特征进行人为提取、提取特征方法不够鲁棒、识别模型参数量大及识别速度慢的缺点,为奶牛个体轻量化识别提供了参考。

In order to solve the problems of large model and slow recognition speed in individual recognition of dairy cows in large-scale intelligent dairy farms,a Multi-Light model was proposed for individual identification of dairy cows.The single cow image was segmented from the complex background of the tagged cow image by using the DeepLab V3 semantic segmentation network.In the Multi-Light model,dilated convolution was introduced to ensure the ability to extract the global information of the image while keeping the number of parameters unchanged.A Multi-scale convolution module was added to enhance the detection ability of the model of feature points at different scales by the model.In the model,shortcut was used to ensure that the features were not lost and the recognition accuracy of the model was improved.In addition,channel attention was used to improve the recognition accuracy of the model and make the model more nonlinear.Finally,the segmented cow image data set was input into the Multi-Light model for training.The results showed that the average recognition accuracy of Multi-Light model for individual cow recognition was 98.51%,which was higher than that of other classical models.Compared with the lightweight model,the size of Multi-Light model was 5.86 mb,and the number of parameters was less on the premise of high recognition accuracy.The results indicated that the lightweight model constructed in this experiment overcame the defects of traditional methods,such as the need for artificial extraction of features,the lack of robustness of feature extraction methods,and the large amount of recognition model parameters and slow recognition speed,and provided a reference for individual lightweight recognition of dairy cows.