摘要
[目的]为解决目前规模化养殖下肉鸡腿病多发、人工检查成本高等问题,提出一种基于红外热图像的肉鸡腿部异常自动检测方法。[方法]根据肉鸡步态评分标准,将肉鸡跛行划分为正常、轻微异常、中度异常、严重异常4个评级;通过YOLO v3神经网络模型获取肉鸡腿部区域的温度数据,并结合OpenCV图像处理方法提取肉鸡身体质心高度、身体前倾角度、身体偏转角度等姿态特征;选用随机森林分类模型进行多特征融合训练,并选取了5种不同分类器进行对比。[结果]随机森林分类模型在准确率、查准率、召回率、F1得分等指标上均占优势,分别为96.16%、96.95%、95.50%和0.965。[结论]该方法可用于肉鸡腿部异常检测,可以有效降低人力成本,为实现白羽肉种鸡疾病早期预防和福利养殖提供技术支撑。
[Objectives]In order to solve the problems of frequent occurrence of leg disease and high cost of manual inspection in large-scale breeding,an automatic detection method of abnormal leg of broiler based on infrared thermal image was proposed.[Methods]According to the gait scoring standard,the claudication of broilers was divided into four grades:normal,slightly abnormal,moderately abnormal and severely abnormal.The temperature data of broiler legs were obtained by YOLO v3 neural network model,and the posture features such as body center of gravity height,body tilt angle and body deflection angle were extracted by OpenCV image processing method,and random forest classification model was selected for multi-feature fusion training.And five different classifiers were selected for comparison.[Results]The random forest classification model had an advantage in accuracy rate,precision rate,recall rate,F1-score and other indicators,which were 96.16%,96.95%,95.50%and 0.965,respectively.[Conclusions]This method can be used to identify abnormal legs of broilers,can effectively reduce labor costs,and provide technical support for early disease prevention and welfare breeding of white-feathered broilers.
作者
许志强
沈明霞
刘龙申
孙玉文
郑荷花
张伟
XU Zhiqiang;SHEN Mingxia;LIU Longshen;SUN Yuwen;ZHENG Hehua;ZHANG Wei(College of Engineering/Jiangsu Key Laboratory of Intelligent Agricultural Equipment,Nanjing Agricultural University,Nanjing 210031,China;New Hope Liuhe Co.,Ltd.,Qingdao 266100,China)
出处
《南京农业大学学报》
CAS
CSCD
北大核心
2021年第2期384-393,共10页
Journal of Nanjing Agricultural University
基金
国家重点研发计划项目(2017YFD0701602)。
