ETHIOPIAN COFFEE LEAF DISEASE DETECTION USING DEEP LEARNING

Abstract

Ethiopia is Africa's biggest coffee exporter, accounting for 22% of total commodity exports. Coffee is an important agricultural crop in the world economy, especially in Ethiopia. Diseases are now recognized by manual assessment by professionals based on visual inspection. However, this presents issues because expertise may not be available in all industrial regions. Additionally, researchers may have difficulty to detect disease such as Cercospora leaf spot. To solve these challenges, the goal of this study is to use digital image processing and deep learning techniques for automated detection of coffee leaf disease. This study uses a dataset of 4000 coffee leaf images from the Jimma and Bonga agricultural Research Centers to identify specific diseases such as coffee leaf rust, Phoma Life Spot, Brown Eye Spot and healthy. The dataset allows for comprehensive training and evaluation of the Convolutional Neural Network (CNN) model, ensuring its effectiveness in accurately identifying different coffee leaf diseases. The CNN model underwent extensive training and refinement to improve its ability to detect subtle patterns and distinguish traits associated with different diseases. The proposed methodology achieved an impressive 95.3% accuracy rate, demonstrating the efficacy and dependability of the CNN-based strategy. This achievement has enormous agricultural uses. A detailed investigation of the CNN's decision-making process provides useful insights into the key qualities required for accurate identification of coffee leaf diseases. Our findings contribute to a better knowledge of how to employ deep learning techniques in complicated contexts, showing CNNs' capacity to successfully solve issues associated with the exact and efficient identification of coffee leaf disease. In future work, we would like to recommend the model's ability to identify patterns in the leaf parts of the coffee plant, including skeletonized patterns, and citrus leaf miner with large images.