AgriSwin: A Swin Transformer-Based Deep Learning Approach for Crop Disease Classification

This research introduces an advanced approach to automated crop disease detection using an Augmented Swin Transformer integrated with Mixup and CutMix augmentation techniques. The study focuses on classifying ten distinct conditions across three major crops—wheat, corn, and potatoes—using a dataset of 10,428 high-resolution images, divided into 70% training, 15% validation, and 15% testing. The dataset includes Brown Rust, Yellow Rust, and Healthy for wheat; Common Rust, Gray Leaf Spot, Northern Leaf Blight, and Healthy for corn; and Early Blight, Late Blight, and Healthy for potatoes. The proposed model enhances feature extraction through self-attention mechanisms, mitigates overfitting via Mixup and CutMix, and improves generalization across diverse agricultural conditions. Experimental results demonstrate superior classification accuracy and robustness against image variations and noise, making it a scalable solution for early disease detection in precision agriculture.