Ensemble of Densely Connected Convolutional Networks for Brinjal Leaf Disease Detection

Abstract

One of the growing concerns in global agriculture is the rise of plant diseases caused by pathogens like viruses, bacteria, and fungi. Detecting these diseases early and implementing effective preventive measures is crucial to limit their spread. Manual disease detection in plants is a time-consuming process. In recent years, deep learning, involving image processing and computer vision, has shown promise. For instance, India's brinjal crop has suffered yield losses due to delayed disease identification. Utilizing automatic disease detection through image processing can assist farmers. While various techniques exist for comparing infected and healthy leaves, the scarcity of diverse datasets containing different leaf diseases has hindered progress in plant disease detection. This paper presents a new dataset and an improved deep learning model based on an ensemble of Densely Connected Convolutional Networks (DenseNet) for brinjal leaf disease detection. Our dataset comprises 8,080 annotated brinjal leaves categorized into seven classes, including six types of diseased leaves and one representing a normal leaf. The enhanced deep learning model is an ensemble of multiple DenseNet architectures designed to enhance overall model performance. Ensemble methods, which combine predictions from multiple models, have been employed to make predictions more accurate and robust compared to individual models. In this paper, we present DenseNet Ensembles, incorporating DenseNet121, DenseNet169, DenseNet201, and DenseNet264. Experimental results demonstrate that the Ensemble DenseNet achieves the highest prediction accuracy at 94.4% when compared to contemporary methods. This advancement represents a significant contribution to the field of agriculture, offering a promising solution to combat plant diseases and improve crop yield sustainability.