Research on Cow Face Recognition Method Based on Improved YOLO v7
高洁 1曹浩2
扫码查看
点击上方二维码区域，可以放大扫码查看
作者信息
1. 安徽科技学院机械工程学院,安徽凤阳 233100
2. 安徽科技学院信息与网络学院,安徽蚌埠 233000
折叠
摘要
基于智能化养殖的发展需求,以牛脸图像作为研究对象,提出了一种基于改进YOLO v7的牛脸识别方法.收集1525张牛脸图像,进行筛选与预处理,形成图像识别数据集.基于目标检测算法YOLO v7的性能基础,针对牛脸识别的特点,提出了适用于牛脸识别的目标检测算法YOLO_C(You Only Look Once_Cattle).首先,使用更适合视觉任务的FReLU激活函数取代了原算法中的SiLU激活函数;其次,将注意力机制CBAM融合到算法网络的骨干层中,提升算法的特征提取能力;最后,在特征融合层引入CARAFE上采样模块以更好恢复牛脸图像细节,提升模型对牛脸个体的识别精度.实验结果表明,YOLO_C与原算法相比,在牛脸识别数据集上识别精确率由87％提升到89.4％,召回率由93％提升到94.4％,平均精度从89％提高到92.9％,检测速度达到83 FPS.
Abstract
Based on the development needs of intelligent farming,taking cow face images as the research object,a cow face rec-ognition method based on improved YOLO v7 is proposed.Collecting 1525 cow face images,and filter and preprocess the ima-ges to form an image recognition dataset.Based on the performance basis of the target detection algorithm YOLO v7,and ai-ming at the characteristics of cow face recognition,a target detection algorithm YOLO_C(You Only Look Once_Cattle)suit-able for cow face recognition is proposed.First,the SiLU activation function in the original algorithm is replaced by the FReLU activation function that is more suitable for visual tasks;Secondly,the attention mechanism CBAM is integrated into the back-bone layer of the algorithm network to improve the feature extraction ability of the algorithm;Finally,the CARAFE upsam-pling module is introduced in the feature fusion layer to better restore the details of the cow face image and improve the recogni-tion accuracy of the model for the individual cow face.The experimental results show that compared with the original algo-rithm,YOLO_C improves the recognition accuracy rate from 87％to 89.4％,the recall rate from 93％to 94.4％,and the av-erage precision from 89％to 92.9％.The speed reaches 83 FPS.
维普
万方数据