Identification Of Tea Leaf’s Diseases In Natural Scene Images Based On Low Shot Learning

Tea trees are susceptible to diseases during the growth process.Tea diseases seriously affect the yield and quality of tea.Effective disease control requires accurate identification of the disease.In recent years,with the development of artificial intelligence and computer vision,it is feasible to use image features to automatically identify plant diseases.However,the tea leaf’s disease image under the natural scene has a complex background and uneven light,which seriously affects the accuracy of disease recognition.At the same time,due to the lack of funds,tea leaf’s disease sample collection is affected by time,weather and collection area,sample types and quantities are very limited.Under the condition of low shot learning,machine learning will produce the problem of overfitting,and the identification accuracy and robustness are insufficient.Therefore,this article focuses on natural scenes and low shot as the premise to study how to improve the accuracy of tea disease recognition.The main research contents and the results of thesis are shown as follows:1.A method for identifying tea leaf’s disease in natural scene images with low shot by using support vector machine and deep network is proposed.In order to solve the problem of complex background in natural scenes,the disease spots are first segmented from the diseased tea leaf’s images.Because the support vector machine is suitable for high-dimensional,high-noise,low shot learning,thesis uses the support vector machine learning method to realize the spots segmentation of tea leaf’s diseases.Secondly,the number of training samples is expanded by means of image augmentation to train deep learning models and solve low shot problems.The segmented tea leaf’s spots image was augmented with an improved conditional deep convolutional generative adversarial networks(CDCGAN-GP),and the expanded tea leaf’s spot image was trained on the VGG16 deep learning classification network to finally realize tea disease recognition.Experiments show that the spots segmentation method based on support vector machine under low shot can effectively improve segmentation accuracy,and the segmented spots image retains the edge information of the spots better.The recognition rate of tea diseases recognition method based onC-DCGAN-GP and VGG16 deep network is higher than that of traditional machine learning methods.2.A method for identifying tea leaf’s diseases in natural scene images based on low shot combined with deep transfer and Cayley-Klein metrics is proposed.After segmenting the disease spots in tea leaf’s disease images using support vector machines,the features of the segmented spots were extracted using deep transfer learning method,and then the classification and recognition of the diseases were realized using Cayley-Klein metric learning.The deep transfer learning model is used to extract the spots features under a low shot,which can avoid the lack of artificial feature extraction and prevent overfitting problems under a low shot.Cayley-Klein metric learning can reflect the spatial structure information and semantic information of samples.Using Cayley-Klein metric learning to identify tea leaf’s diseases can effectively improve the recognition accuracy.The experimental results show that compared with classic machine learning methods and deep learning methods,the method of tea leaf’s disease recognition under a low shot combining deep transfer and metric learning has a higher average recognition accuracy.