A Two-Stage Learning Framework for Coffee Disease Detection

Abstract

Coffee production is a key component of the agri-cultural economy of the Central Highlands, with considerable commercial value derived from the Robusta and Arabica varieties grown in a variety of ecological zones. However, coffee production is hampered by geographically distributed disease outbreaks that require sophisticated, real-time monitoring capabilities, and has not benefited from modern technological advances in digitization and artificial intelligence. The recent climate changes cause geographically distributed disease out-breaks - require more sophisticated monitoring mechanisms. Existing disease detection and response systems are severely constrained by the lack of data analytics tools to provide timely, location-aware information. To fill the critical gap, this paper will create an integrated deep learning framework to detect coffee diseases with computer vision techniques and support real-time disease surveillance and management decisions. The proposed two-stage learning framework could classify diseases into seven categories: rust, pink disease, maly-bug infestation, nematode damage, Phoma leaf spot, leaf miner damage and healthy foliage. YOLO is used to precisely locate the symptom areas. By integrating the automated detection module into the digital mapping platform, the prevalence of the disease can be monitored spatiotemporal manner, enabling early warning systems and targeted intervention strategies for sustainable coffee production in the region.