<![CDATA[Precision disease detection in citrus commodities has become increasingly essential within the framework of Agriculture 4.0, particularly for small-scale vendors who still rely on manual visual inspection that is often inconsistent and error-prone. This study develops and evaluates a web-based citrus fruit disease detection system using the MobileNet V2 Convolutional Neural Network architecture. The methodological novelty of this work lies in the integration of an optimized MobileNet V2 model enhanced through targeted data augmentation and lightweight fine-tuning into an end-to-end web ecosystem that supports two inference modes: static image upload and real-time webcam-based detection, tailored to the operational needs of small vendors. The system classifies citrus fruit images into four categories: Black-Spot, Citrus Canker, Greening (Huanglongbing), and Fresh, using more than 1,000 augmented images standardized to 224×224 pixels with an 80:20 train–test split. Experimental results show that the model achieves an accuracy of 96.21%, with consistently high precision and recall across disease classes, while the Fresh class exhibits relatively higher misclassification due to visual similarity with early-stage symptoms. The Flask-based web application demonstrates stable performance under black-box testing and delivers rapid, high-confidence predictions. These findings affirm the effectiveness of lightweight CNN approaches in improving fruit quality inspection accuracy, reducing sorting errors, and supporting more efficient workflows for local vendors.]]>
