Deep Learning for Cotton Disease Detection Lightweight, Explainable and Field-Ready Solutions

Abstract

Cotton is a crucial crop for the economy that is globally recognized as white gold and a major contributor to the Indian economy. However, cotton production is threatened due to various diseases affecting the leaves, like bacterial blight, leaf curl virus, fungal infections and pest attacks impacting the crop yield and quality that affect economic losses. The traditional disease detection methods, which depend on manual inspection, are inefficient, time-consuming, laborious, inaccurate and lead to misdiagnosis and often unreliable under field conditions. The need for early and accurate diagnosis is critical for timely intervention. In recent years Machine Learning (ML) and Deep Learning (DL) have been used for automated disease detection through leaf images. The study highlighted lightweight convolutional neural networks (CNNs), transformer-based models and object detection frameworks such as YOLO and RT-DETR, which have performed accurate results. Transfer learning with advanced backbones (EfficientNet, Xception, ResNet), integrating with attention mechanisms (e.g., CBAM, DFSA) for feature enrichment. The Explainable AI (XAI) for improving the explainability, while synthetic data generation using GANs reduces dataset imbalance. The review consolidated the current state of DL models for cotton disease detection, focusing on optimized approaches for mobile and edge deployment. Finally, it identifies existing research gaps and future directions for accurate, efficient, and field‑ready solutions.