Research Of Tea Leaf Disease Images Recognition Based On Capsule Network

Tea is an important economic crop.China is the largest tea planting area in the world,and its disease directly affects the yield and quality of tea.Traditional identification methods not only require tea farmers to have expertise disease,but also time-consuming and laborintensive,and accuracy is low.In recent years,Convolutional Neural Network has been used widely in crop disease image recognition and achieved good results.However,Convolutional Neural Network does not consider the spatial relationship between features,which has some limitations,while Capsule Network uses vectors to represent the relative relationship of entities in the image,which can fully mine the relationship information between the disease image features.This thesis takes tea leaf disease image as the research object,and uses Capsule Network to carry out the research work on the image enhancement and recognition of tea leaf disease.The main research contents are as follows:(1)In view of the low occurrence of some tea leaf diseases and difficulty in image acquisition,an image enhancement algorithm for tea leaf diseases based on a Generative Adversarial Capsule Network was proposed.Generative Adversarial Network is an effective method for data enhancement,but it has problems of instability and poor quality of generated images.Capsule network can better fit features and improve the quality of generated images,but the growth rate of squash function in dynamic routing is slow,so the squash function is improved,and improved Capsule Network is used as a discriminator in the Generative Adversarial Network to generate tea leaf disease images.At the same time,in order to solve the problem of unstable training of the Generative Adversarial Network,the Wasserstein distance loss function is used in the training.Experimental results show that the proposed algorithm solves the problem of unstable training and can generate high-quality tea leaf disease images.(2)Aiming at the problem of the high similarity of different tea leaf diseases and how to extract disease effectively,a method of tea leaf disease image recognition based on Capsule Network with attention mechanism was proposed.The attention mechanism was introduced into Capsule Network to extract the effective and detailed feature.In response to the problem that the attention mechanism does not consider the local correlation between channels,the input feature channels of channel attention module are grouped to improve the feature expression ability.At the same time,residual dense blocks are used to extract hierarchical feature.The experimental results show that the proposed Capsule Network with attention mechanism effectively improves the accuracy of tea leaf diseases image recognition.(3)Aiming at the problem that Capsule Network has only one convolutional layer and extraction of deep abstract features is insufficient,a method based on Capsule Network with Res2 Net for tea leaf disease recognition was proposed.As a deep Convolutional Neural Network,Res2 Net can not only extract deep abstract features,but also express multi-scale features.The features of tea leaf disease are shown in different size areas,and multi-scale receptive fields are required to capture features.The improved Res2 Net is introduced into the Capsule Network,which improves the ability of feature extraction.The experimental results on the tea leaf disease image dataset verify that the proposed method is better than Capsule Network and other methods.