Multiple deep learning system for robust Wheat leaf disease severity estimation

The Yellow Rust is a pathogen that affects wheat leaves causing significant losses and a gradual decline in wheat worldwide quality. This disease causes deformations in the shape and texture of wheat leaves. Recent advances in image analysis reached remarkable capability of recognizing patterns, in particular, Convolutional Neural Networks (CNN). In this paper, a Yellow Rust disease severity level estimation system based on the combination of various CNN models is proposed. Precisely, VGG16, MobileNetV2, InceptionV3, Yellow-Rust-Xception and ResNet152V2, are trained as end-to-end systems that estimate six severity levels of the Yellow Rust disease. Then, deep features are extracted from fully connected layers of each model and undergo a feature selection and concatenation processing. Derived deep features are finally trained by a multilayer perceptron to perform the classification into various severity levels. Experiments are conducted on the Yellow-Rust-19 dataset, which contains 15000 wheat leaf images. Results obtained illustrate that effectiveness of the proposed system which outperform bot individual CNN and various state of the art systems.