Automated identification and classification of Coffee Leaf Diseases using a Deep Learning Approach

Abstract

Coffee is regarded as the most consumed drink around the globe and has accounted for a major source of income in the regions where it is cultivated. To meet the coffee marketplace's requirements around the globe, cultivators must improve and analyze its cultivation and quality. Several factors, such as environmental changes and plant leaf diseases such as Miner, Rust, Phoma, and Cercospora are major hindrances to increasing the yield of coffee. These diseases often appear visually similar in their early stages, making accurate identification challenging using traditional methods. To overcome this issue, we developed a deep learning solution based on 2D Convolutional Neural Networks (2D-CNN) to automatically detect and classify these diseases from leaf images. Our dataset was compiled from three publicly available sources and enhanced through LAB color space transformation and Contrast Limited Adaptive Histogram Equalization (CLAHE) to improve image clarity. We implemented a multi-stage classification approach, tailoring the CNN architecture and data augmentation strategies at each stage to maximize performance. The model achieved a remarkable 99.86% accuracy on the final test set, demonstrating its strong potential as a reliable and efficient tool for disease detection in real-world coffee farming scenarios.