Development and Validation of a Deep Learning Algorithm for the Recognition of Plant Disease

Abstract

Crop health is the foundation for agricultural development. In some area, due to the lack of professional botanical experts, it is difficult to correctly diagnose the plant disease in the work of planting. In this paper, we propose a novel deep neural network structure that can reliably classify plant types and plant disease using a single image of plant leaf which enables the end-to-end diagnosis of plant disease. Our proposed model consists of two sub-models, a leaf segmentation model that employs a U-Net to separate the leaves in the original image from the background which effectively eliminates interference, and a plant disease classification model based on our proposed Two-head network that classifies plant diseases with the features extracted by various popular pre-trained models. We verified our model using the plant disease dataset with 8 plant species and 19 plant diseases provided by AI Challenger 2019. Experimental results demonstrate that our final model achieves a 0.9807 accuracy of plant classification and a 0.8745 accuracy of disease recognition. We believe that the technology in our model has great potential to become the basis of fully automatic reliable plant disease classification system which can be embedded into portable devices to assist farmers in the future.