引用本文:杨君艳,孙瑞志,靳晨鹏,尹宝全.基于多尺度卷积的蛋鸡肠道疾病识别方法研究[J].中国农业信息,2022,34(6):14-26
【打印本页】   【HTML】   【下载PDF全文】   查看/发表评论  【EndNote】   【RefMan】   【BibTex】
←前一篇|后一篇→ 过刊浏览    高级检索
本文已被:浏览 292次   下载 79 本文二维码信息
码上扫一扫!
分享到: 微信 更多
基于多尺度卷积的蛋鸡肠道疾病识别方法研究
杨君艳1,孙瑞志1,2,靳晨鹏1,尹宝全2
1.中国农业大学信息与电气工程学院,北京 100083;2.中国农业大学烟台研究院,山东 烟台 264670
摘要:
【目的】 在蛋鸡养殖中,如球虫病和传染性法氏囊病等鸡肠道疾病发病率高,严重影响鸡舍饲料的利用率,增加了养殖成本。为在密集的蛋鸡笼养环境中使用肠道疾病智能化诊断方法,实现多种肠道疾病的分类识别,【方法】 文章根据病理粪便的差异,提出了基于多尺度卷积的对鸡肠道疾病进行分类的算法。首先,实验共采集正常粪便、稀便、绿便和血便4类图像共1 834幅,建立蛋鸡粪便数据集。为减少过拟合的出现,对数据集进行了数据增强,扩充数据集至5 128幅。其次,对VGG16模型进行改进,使用全局平均池化代替全连接层,同时引入了3×3和5×5的多尺度卷积和通道注意力机制,将SE模块插入到多尺度卷积的末端,构建了一个基于多尺度卷积的蛋鸡肠道疾病识别网络VGG-MSC。【结果】 VGG-MSC的各项指标较VGG16均得到提高,其中在蛋鸡粪便数据集上的分类准确率达到了98.04%,较VGG16提高了1.75%,可为鸡的肠道疾病诊断提供有效的决策支持。【结论】 以深度学习为基础,该方法能够实现对鸡肠道疾病的智能诊断,有利于及早发现并预防鸡肠道疾病,为畜禽养殖提供信息化服务,推动蛋鸡产业持续高质量发展。
关键词:  深度学习  计算机视觉  卷积神经网络  图像识别  疾病诊断
DOI:10.12105/j.issn.1672-0423.20220602
分类号:
基金项目:国家重点研发计划项目“黄羽肉鸡表型的智能化精准测定技术研发”(2021YFD13001001)
Research on the identification method of intestinal diseases in laying hens based on multi-scale convolution
Yang Junyan1, Sun Ruizhi1,2, Jin Chenpeng1, Yin Baoquan2
1.College of Information and Electrical Engineering,China Agricultural University,Beijing 100083,China;2.Yantai Institute,China Agricultural University,Yantai Shandong 264670,China
Abstract:
Purpose In the breeding of laying hens,the incidence of intestinal diseases such as coccidiosis and infectious bursal disease is high,which seriously affects the utilization rate of feed in the henho use and increases the breeding cost. In order to use the intelligent diagnosis method of intestinal diseases in the intensive laying hens cage environment and realise the classification and identification of various intestinal diseases,Method This paper proposes an algorithm for classifying chicken intestinal diseases based on multi-scale convolution according to the difference of pathological feces. First,a total of 1 834 images of 4 types of normal feces,loose stools,green stools and bloody stools were collected in the experiment to establish a feces dataset of laying hens. In order to reduce the occurrence of overfitting,the data set is enhanced and the data set is expanded to 5 128. Second,improved the neural network VGG16 - the global average pooling layer was used to instead of the fully connected layer,at the same time,the multi-scale convolutions of 3×3 and 5×5 and channel attention mechanism were introduced,and the SE module was inserted into the end of the multi-scale convolution. The new convolutional neural network was named VGG-MSC,which structure was based on multi-scale convolution.Result The indicators of VGG-MSC were improved compared with those of VGG16,and the classification accuracy of VGG-MSC on the egg chicken manure dataset reached 98.04%,which was 1.75% higher than that of VGG16. This method can provide effective decision support for the diagnosis of intestinal diseases in chickens.Conclusion Based on deep learning,this method can realize intelligent diagnosis of chicken intestinal diseases,which is conducive to early detection and prevention of chicken intestinal diseases,provide information services for livestock and poultry breeding,and promote the sustainable and high-quality development of laying hens industry.
Key words:  deep learning  computer vision  convolutional neural networks  image recognition  disease diagnosis