A low shot learning method for tea leaf’s disease identification

Abstract

Tea leaf’s diseases seriously affect the yield and quality of tea. This paper presents a low shot learning method for tea leaf’s disease identification in order to prevent and control tea leaf’s diseases timely. By extracting the color and texture features, disease spots on tea leaf's disease images are segmented by using support vector machine (SVM) method. With segmented disease spot images as input, new training samples are generated by the improved conditional deep convolutional generative adversarial networks (C-DCGAN) for data augmentation, which are used to train VGG16 deep learning model to identify the tea leaf's diseases. Experimental results show that, SVM can segment disease spot images on the condition of low shot learning while retaining the edge information well, improved C-DCGAN can generate augmented images with the same data distribution as real disease spot images, the VGG16 deep learning model trained with augmented disease spot images can identify tea leaf’s diseases accurately, and the average identification accuracy of the proposed method reaches 90%, far exceeding classical low shot learning methods.