基于计算机视觉的笼内蛋鸡体重测量  

Weight measurement of caged laying hens based on computer vision

作　　者：李可强 LI Keqiang(College of Mathematics and Information Science&Technology,Hebei Normal University of Science&Technology,Qinhuangdao 066004,China;Hebei Innovation Center for Smart Perception and Applied Technology of Agricultural Data,Qinhuangdao 066004,China;Hebei Key Laboratory of Special Animal Germplasm Resources Mining and Innovation,Qinhuangdao 066004,China)

机构地区：[1]河北科技师范学院数学与信息科技学院,河北秦皇岛066004 [2]河北省农业数据智能感知与应用技术创新中心,河北秦皇岛066004 [3]河北省特色动物种质资源挖掘与创新重点实验室,河北秦皇岛066004

出　　处：《黑龙江畜牧兽医》2024年第20期38-44,124,125,共9页Heilongjiang Animal Science And veterinary Medicine

基　　金：河北省鸡现代种业科技创新团队子项目“鸡智能育种技术研究”(21326303D-7);河北省农业数据智能感知与应用技术创新中心开放基金项目(ADIC2023Y007)。

摘　　要：为了解决规模化蛋鸡养殖中传统称重法极易引发蛋鸡应激反应、影响蛋鸡生产性能和鸡蛋品质的问题,试验采用一种基于计算机视觉的笼内蛋鸡体重测量方法,首先通过深度相机在不同距离、不同角度采集鸡笼中蛋鸡图像,克服距离、拍摄角度对测量精度产生的影响;其次利用YOLOv5s模型检测蛋鸡躯干,减少背景对特征提取的干扰;最后通过拼接卷积注意力模块与Resnet50神经网络提出两种测量方案。结果表明:两种测量方案在平均相对误差和平均绝对误差上均优于Resnet50神经网络模型;在相机距蛋鸡40~80 cm时的测量拟合度均大于0.994。说明本研究提出的蛋鸡体重测量方法的精度和相关系数均能满足实际生产需求,可为蛋鸡体重测量提供新的理论参考。In order to solve the problem that the traditional weighing method can easily cause stress reaction of laying hens in large-scale breeding,which affects production performance and egg quality of laying hens,in this experiment,a method for measuring the weight of caged laying hens based on computer vision was used to collect images of hens in a cage by depth cameras at different distances and angles,so as to overcome the influence of distance and shooting angle on measurement accuracy.Then,the YOLOv5s model was used to detect the torso of laying hens to reduce the interference of background on feature extraction.Finally,two measurement models were proposed by splicing the convolutional attention module with Resnet50 neural network in this study.The results showed that the average relative eror and average absolute error obtained by both measurement schemes were better than the Resnet50 neural network model.When the camera was 40~80 cm away from the laying hens,the fitting degree of the measurement was greater than 0.994.The results indicated that the weight measurement models proposed in this study could meet the actual production requirements in terms of accuracy and correlation coefficients,and could provide a newreference forthe weight measurement of laying hens.

关 键 词：蛋鸡 鸡笼 体重测量 计算机视觉 神经网络 

分 类 号：S818.9[农业科学—畜牧学] S815.5[农业科学—畜牧兽医]