Automated Identification of Crop Diseases using Computer Vision

Early and accurate identification of crop diseases is essential for ensuring agricultural productivity, food security, and sustainable farming practices. This study presents an automated computer vision based framework for multi-crop disease classification using a lightweight deep learning architecture. The proposed system employs the ReXNet-1.5 convolutional neural network as the core feature extractor, integrating efficient hierarchical feature learning with low computational complexity. A publicly available multi-crop dataset comprising 13,324 images across 17 disease and healthy classes covering corn, rice, potato, wheat, and sugarcane is used for model training and evaluation. Experimental results demonstrate strong performance, achieving 97.45% accuracy and a macro-F1 score of 96.26%, indicating reliable class-balanced prediction under dataset imbalance. Grad-CAM based visual explainability is incorporated to provide interpretable disease localization, enhancing transparency and trust in model predictions. Additionally, the model exhibits high computational efficiency, enabling real-time inference suitable for deployment on resource constrained platforms. The proposed framework offers an accurate, interpretable, and deployable solution for real world crop disease diagnosis, supporting intelligent decision-making and scalable agricultural monitoring systems.