A DEEP LEARNING ENABLED FRAMEWORK FOR AUTOMATED CASSAVA CROP DISEASE DETECTION

Abstract

Cassava is a major staple crop in sub-Saharan Africa and plays a crucial role in food security and rural livelihoods, but its productivity is significantly affected by plant diseases that are often difficult to detect at early stages using traditional visual inspection methods. These conventional approaches are typically slow, subjective, and unreliable, especially in resource-limited farming environments. This study presents the development of a deep learning-based cassava crop disease detection system aimed at improving early diagnosis and supporting effective cassava crop management. The system is built using a Convolutional Neural Network (CNN) enhanced with transfer learning to enable efficient extraction of discriminative features from cassava leaf images. A hybrid development methodology combining Agile, DevOps, and Rapid Application Development (RAD) principles was adopted to support iterative development, continuous testing, and user-centered improvements involving agronomists and farmers. The model was trained using a combined dataset of field-collected images and publicly available plant disease repositories covering five classes: Cassava Mosaic Disease (CMD), which stands for Cassava Mosaic Disease, Cassava Brown Streak Disease (CBSD), which stands for Cassava Brown Streak Disease, Cassava Green Mite (CGM), Cassava Bacterial Blight (CBB), and healthy cassava leaves. To enhance robustness under real-world conditions, preprocessing techniques such as resizing and normalization were applied, alongside data augmentation methods including rotation, flipping, and brightness adjustments. The system was evaluated using accuracy, precision, recall, F1-score, and confusion matrix analysis, achieving an overall accuracy of 75%. Strong performance was observed in detecting CMD and CBSD, while lower performance occurred in distinguishing healthy leaves and early-stage infections due to visual similarity between classes. Overall, the results demonstrate that CNN-based approaches are effective for cassava disease detection and can support farmers in early diagnosis, reduce crop losses, and improve agricultural productivity.