SmartLeafNet: An Intelligent Swin Transformer-Based Model for Crop Disease Detection

Abstract

Agriculture plays a vital role in the economy of Andhra Pradesh, with over 60% of the population relying on it either directly or indirectly for their livelihood. However, crop diseases continue to pose serious challenges, leading to an estimated 20–30% loss in yield annually. Among the major crops cultivated in the region, corn, potato, and tomato are highly susceptible to various leaf-based diseases caused by fungal, bacterial, and viral infections. These issues significantly impact productivity, farmer income, and food security. Early and accurate detection of diseases is therefore essential to reduce crop losses and limit excessive pesticide usage. Traditionally, Crop Disease Detection (CDD) has relied on manual inspection, which is time-consuming, subjective, and prone to human error, making it unsuitable for large-scale farming. To address these limitations, this study proposes an image-based Crop Disease Detection Network (CDD-Net) using leaf images. The system begins with preprocessing techniques such as resizing and normalization to ensure consistency in input data. Feature extraction is performed using the Swin Transformer, which captures rich spatial and contextual information. For performance evaluation, models such as Logistic Regression Classifier (LRC), eXtreme Gradient Boosting (XGB), and Gaussian Naïve Bayes Classifier (GNBC) are implemented. Finally, a Stochastic Gradient Descent (SGD) Classifier is proposed to improve accuracy and robustness. Experimental results demonstrate that the proposed approach enhances disease detection performance, providing a reliable and scalable solution for precision agriculture.