基于卷积胶囊网络的百合病害识别研究  

Disease Detection of Lily Based on Convolutional Capsule Network

作　　者：丁永军 张晶晶 李民赞[3] DING Yongjun;ZHANG Jingjing;LI Minzan(School of Electronic and Information Engineering,Lanzhou City University,Lanzhou 730070,China;College of Computer Science and Engineering,Northwest Normal University,Lanzhou 730070,China;Key Laboratory of Modern Precision Agriculture System Integration Research,Ministry of Education,China Agricultural University,Beijing 100083,China)

机构地区：[1]兰州城市学院电子与信息工程学院,兰州730070 [2]西北师范大学计算机科学与工程学院,兰州730070 [3]中国农业大学现代精细农业系统集成研究教育部重点实验室,北京100083

出　　处：《农业机械学报》2020年第12期246-251,331,共7页Transactions of the Chinese Society for Agricultural Machinery

基　　金：国家自然科学基金项目(31971785);甘肃省自然科学基金项目(18JR3RA224);甘肃省社科规划项目(YB087)。

摘　　要：为了提高百合病害诊断模型的抗噪能力,以VGG16模型为基础构建卷积胶囊网络,并分析了胶囊尺寸、路由迭代次数对训练时间及模型精度的影响。最终得到胶囊尺寸为8、路由迭代次数为3的卷积胶囊网络,该网络对百合病害诊断精度达到99.20%。使用不同等级的高斯噪声、椒盐噪声、斑点噪声、仿射变换图像对模型抗噪能力进行测试,结果表明,卷积胶囊网络明显优于VGG16模型,更适合在实际生产环境下的百合病害诊断。Lanzhou lily is the only kind of sweet lily in China and it is one of the famous specialties of Gansu Province.However,its yield and quality were decreased significantly in recent years due to gray mold disease,bulb rot disease and other diseases and insect pests.In order to improve the anti-interference ability of Lanzhou lily diseases diagnosis model,the three full connection layers of VGG16 convolutional network was replaced with capsule network module to construct convolutional capsule network.And the effects of capsule size and route iteration times on training time and model accuracy were analyzed systematically.The result of the experiment showed that the diagnosis accuracy of Lanzhou lily diseases via convolutional capsule network was 99.20%when the capsule size was 8 and the route iteration time was 3.And the capsule size and the number of routing iterations had no significant effect on the accuracy of the model.In addition,the accuracy of VGG16 model was slightly higher than that of convolutional capsule network when the affine transformation grade was 0.04~0.08.But the anti-interference ability of convolutional capsule network was obviously better than that of VGG16 model for Gaussian noise,salt-and-pepper noise,speckle noise and other grades of affine transformation.So it was possible to use the convolutional capsule network for dealing with the real-world examples of Lanzhou lily diseases recognition.

关 键 词：百合 病害 识别 卷积胶囊网络 

分 类 号：S644.1[农业科学—蔬菜学] TP391.4[农业科学—园艺学]