Published Paper Details:

Rubber (Hevea brasiliensis) is a commercially important crop, yet foliar diseases such as Anthracnose, Corynespora Leaf Fall, and Phytophthora-induced Dry Leaf Syndrome significantly reduce plantation yield and tree longevity. Manual diagnosis by agricultural officers is time-consuming, subjective, and often inaccessible to smallholder farmers in remote regions. This paper presents an end-to-end AI-powered mobile application for real-time detection and classification of rubber leaf diseases using deep learning. The system employs a MobileNetV2 convolutional neural network trained via a two-stage transfer learning strategy: an initial warmup phase with a frozen ImageNet backbone followed by fine-tuning of the top 30 layers using cosine-decay learning rate scheduling. Data augmentation, L2 regularization, dropout, and class-weight balancing were applied to improve generalization on the Mendeley Rubber Leaf Dataset containing 1,741 images across four classes. The model achieved a validation accuracy of 99.7% with perfect precision and recall across all classes. The trained model was deployed through a dual-inference architecture: a Python Flask REST API for cloud-based predictions and a quantized TensorFlow Lite model embedded within a cross-platform Flutter mobile application for offline inference, ensuring uninterrupted functionality in rural areas with limited connectivity. The application further incorporates a 90% confidence-threshold gating mechanism to reject ambiguous inputs, multilingual support, an analytics dashboard for disease trend monitoring, and PDF report generation for agricultural record-keeping. The proposed system demonstrates that lightweight deep learning models, combined with intelligent mobile deployment strategies, can provide scalable, accurate, and accessible diagnostic tools for precision agriculture. Index Terms--Deep Learning, Transfer Learning, MobileNetV2, Rubber Leaf Disease, TensorFlow Lite, Flutter, Precision Agriculture, CNN, Image Classification