An Image-based Approach for Early Detection and Classification of Crop Leaf Diseases

Early and accurate detection of plant diseases is vital for sustainable agriculture and minimizing crop loss. This paper presents an image-based system that leverages deep learning techniques to identify and classify plant diseases using leaf images. The proposed model employs a lightweight, custom Convolutional Neural Network (CNN) architecture with transfer learning, achieving high classification accuracy while ensuring real-time inference capability. The dataset is structured into 30 disease classes and undergoes preprocessing steps such as resizing, normalization, and one-hot encoding to improve model performance. The system is further evaluated using confusion matrices and F1-confidence curves, providing deep insights into per-class performance. In addition to automated prediction, the system introduces a novel Image Analysis module that enables users to visualize uploaded images through transformations such as grayscale, thresholding, edge detection, and inversion. This enhances user understanding, supports manual inspection, and builds trust in AI-driven predictions. Results indicate a top classification accuracy of 97.5% and fast inference time, making the system suitable for mobile deployment in field conditions. The work contributes toward scalable, interpretable, and user-friendly plant disease diagnosis, with potential to improve agricultural resilience, especially in low-resource and rural environments.