基于多任务学习的番茄叶片图像病害程度分类

针对现有的分类深度神经网络大多为扁平型的网 络架构,很少关注数据类别的层次 性结构,导致分类器训练难度较大的问题,本文提出一种基于数据层次关系的多任务学习分 类网络模型。依托番茄叶片病害的层次结构信息设计了一个带有共享网络的由粗粒度到细粒 度的层次结构进行病害程度分类,网络模型以ResNet-50作为网络主干,包括两个子网络: 粗粒度网络模块负责区分番茄病害共5类,细粒度网络模块在残差网络模块的基础上添加 SE模块负责病害程度的分类共9类。通过对网络架构各个分支的验证,以及同VGG-16、 ResNet-34、ResNet-50 3种扁平型网络在病害程度分类任务上做比较,证明本文网络结构 的 可行性和有效性,最终测试集分类精度达到93.97%。证明本文结合数 据与网络的层次结构 采用多任务分类方法,是一种有效的病害程度分类算法。

Classification of tomato leaf image disease degree based on multi-task learning

In order to solve the problem that most of the existing deep neural ne tworks for classification are flat network architectures and seldom pay attention to the hi erarchical structure of data categories,which leads to the difficulty of classifier training,a multi- task learning classification network model based on hierarchical relationship of data is proposed.Relying on tomato leaf disease in the hierarchical structure of the information design with a Shared network co mposed of coarse- grained to fine-grained hierarchy for the degree of disease classification,a n etwork model with ResNet-50as the backbone network,including two sub networks:coarse-graine d network module is responsible for the distinction between tomato diseases,a total of five clas ses,fine-grained network module in the residual network module on the basis of adding SE module i s responsible for the degree of disease classification,a total of 9class.Through the verific ation of each branch of the network architecture and the comparison with the three flat networks of VGG -16,ResNet-34and ResNet-50in the classification task of disease degree,the feasibility and effectiveness of the network structure in this paper are proved,and the classification accuracy of t he final test set reaches 93.97%.It is proved that the multi-task classification method combined with the hierarchical structure of data and network in this paper is an effective algorithm for disea se degree classification.