Groundnut Leaf Disease Classification Using Deep Transfer Learning

Abstract

Groundnut is an essential crop that substantially impacts global nutrition, economy, and soil health via nitrogen fixation. Nevertheless, leaf diseases such as early and late leaf spot, Alternaria leaf spot, and rust cause significant output reductions, presenting a considerable challenge to groundnut cultivation. Timely identification of these diseases is crucial for efficient management; however, conventional procedures are labour-intensive, prone to errors, and impracticable for extensive implementation. Convolutional Neural Networks (CNNs) have exhibited exceptional efficacy in image classification across various domains, including healthcare, robotics, and agriculture. Nonetheless, these deep learning techniques are computationally intensive and demand significant memory and power resources. This study utilises transfer learning to address these issues, employing three pretrained CNN models, namely, ResNet50, InceptionV3, and DenseNet201, for the leaf disease classification of groundnut. A meticulously curated collection of groundnut leaf images was utilised, and comprehensive experiments were accomplished to evaluate the models’ efficacy. Out of the three models, DenseNet201 attained the highest performance, achieving an accuracy of $99.89 \%$. This indicates the model’s ability to deliver precise and scalable solutions for early disease detection, offering farmers an affordable resource to improve crop yields and support sustainable agricultural methods.