羊面部图像自动采集系统设计  

Design of Automatic Sheep Facial Image Acquisition System

作　　者：刘苏慧 LIU Suhui(Inner Mongolia Agricultural University,Hohhot Inner Mongolia 010018,China)

机构地区：[1]内蒙古农业大学,内蒙古呼和浩特010018

出　　处：《现代畜牧科技》2023年第12期142-144,共3页Modern Animal Husbandry Science & Technology

基　　金：内蒙古科技计划项目《大型养殖场的羊面部身份识别关键技术研究及应用示范》(2021GG0111)。

摘　　要：如今越来越倾向于智慧养殖,羊脸图像采集具有非接触和对羊无应激的优势,因此通过羊脸图像识别来实现羊身份管理具有重要的实践意义,而羊身份信息的采集是实现羊个体身份识别的重要前提。该文探讨如何快速高效的采集高质量的羊面部图片,分析YOLOv5目标检测算法的网络结构和目标的预测过程,从多尺度特征融合的角度,对YOLOv5网络进行改进,构建新的目标检测算法DFD-YOLOv5,检测速度提升12%左右,再将Alexnet网络模型通过堆叠的方式,将一个多分类问题转换为多个二分类问题从而获对图像的清晰度进行筛选,将模糊的羊面部图片进行移除。Nowadays,more and more people are inclined to smart farming.Sheep face image acquisition has the advantages of non-contact and non-stress to the sheep.Therefore,it is of great practical significance to realize sheep identity management through sheep face image recognition,and the collection of sheep identity information is an important prerequisite to realize sheep individual identity recognition.This paper discusses how to quickly and efficiently collect high-quality sheep face images,analyzes the network structure of YOLOv5 target detection algorithm and the target prediction process,improves the YOLOv5 network from the perspective of multi-scale feature fusion,and constructs a new target detection algorithm DFD-YOLOv5.The detection speed is increased by about 12%.Then,the Alexnet network model was stacked to convert a multi-classification problem into multiple binary classification problems so as to filter the clarity of the image,and the fuzzy sheep face pictures were removed.

关 键 词：数据采集 羊脸识别 目标检测 畜牧业 YOLOv5 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] S826[自动化与计算机技术—计算机科学与技术]