Dense convolutional neural networks based multiclass plant disease detection and classification using leaf images

Abstract

Plant diseases are one of the major concern in the agricultural domain and their automatic identification is very crucial in monitoring the plants. Most of the disease symptoms are reflected in the leaves of plants but the leaf diagnosis by experts in laboratories are costly and time-consuming. In this paper, a deep-learning-based approach is presented for the plant disease detection and classification from leave images captured in various resolutions. Dense convolutional neural network architecture is trained on a large plant leaves image dataset from multiple countries. Six crops in 27 different categories are considered in the proposed work in laboratory and on-field conditions. Images have several inter-class and intra-class variations with complex and challenging conditions that have been addressed in this dense neural network. Five-fold cross-validation and testing on unseen data is done for exhaustive evaluation of the trained model in various parameters. Experimental results proved that the proposed deep learning-based system can efficiently classify various types of plant leaves with good accuracy. The experimental findings demonstrate that an average cross-validation accuracy of 99.58% and average test accuracy of 99.199% is obtained on unseen images with complex background conditions. The processing time to process a single plant leaf image is 0.016 s with significant accuracy which signifies its real-time performance.