Edge AI Based Real-Time Arecanut Plant Disease Detection Using Deep Learning and Smartphone Deployment

Arecanut (Areca catechu) is an important plantation crop in India, and identifying diseases at an early stage can be difficult in the field. This paper presents an Android application for preliminary arecanut disease detection using deep learning and on-device inference. A MobileNetV2 model converted to TensorFlow Lite classifies accepted arecanut images using meaningful class names such as Bud borer, Healthy Leaf, Stem bleeding, and Yellow leaf disease rather than displaying numeric class indices. A separate input gate checks whether an image is related to arecanut before disease classification. The application displays the predicted class, confidence score, symptoms, possible causes, remedies, and pesticide-related information. It also includes an offline Farmer Assistant that answers a limited set of predefined farming questions, such as what causes a detected disease, its symptoms, and basic management guidance, without requiring an internet connection or an online AI service. The reported evaluation includes 2,032 test images after exact duplicate images were excluded. The Keras model achieved 99.31% accuracy, while standard TFLite and Float16 TFLite achieved 98.97% and 99.21%, respectively. The Float16 model reduced the reported model size from 8.50 MB to 4.28 MB. Functional testing recorded 11 successful cases out of 14 for the existing detection workflow; the offline Farmer Assistant is described as a limited, rule-based guidance feature and its performance is not included in these model accuracy figures. Further field testing is required.