PD2SE-Net: Computer-assisted plant disease diagnosis and severity estimation network

Abstract

The plant disease diagnosis and severity estimation are a very challenging research field in the agriculture sector. In this work, we present a robust image-based Plant Disease Diagnosis and Severity Estimation Network (PD2SE-Net), which contains a residual structure and shuffle units. The aim of this paper is to design a more excellent and practical diagnosis system for plant diseases. The common plant disease diagnosis, disease severity estimation, are simultaneously addressed by the proposed PD2SE-Net. In addition, the data augmentation and visualization of convolutional neural networks (CNNs) are exploited in this paper to improve the accuracy and accelerate the excellent selection of hyper-parameters during the training period. To the best of our knowledge, this report for the first time describes a computer-assisted approach that can simultaneously estimate disease severity, recognize species, and classify disease for plants base on deep learning. The proposed PD2SE-Net50 consists of the ResNet50 architecture as the basic model and shuffle units as the auxiliary structures, and it achieves excellent comprehensive performances (overall accuracies of 0.91, 0.99 and 0.98 for the disease severity estimation, plant species recognition and plant disease classification, respectively) over the existing approaches. As a diagnosis expert, our system exploits the multivariate nature of plant leaves in order to deliver an outstanding classification performance with a low computational cost. The experimental results demonstrate the feasibility and effectiveness of our network.