IoT-Enabled MobileNetV3-Tiny with CBAM for Real-Time Sweet Lemon Leaf Disease Detection

Sweet lemon (Citrus limetta) production is significantly affected by leaf diseases such as citrus canker, greening (Huanglongbing), and blackspot, leadingtosevere yield losses and economic impact for farmers. Traditional manual inspection methods are time-consuming, error-prone, and often result in delayedinterventions. In this work, we present an IoT-enabled real-time plant disease detection system using a lightweight MobileNetV3-Tiny model enhancedwithConvolutional Block Attention Module (CBAM). The proposed model was trained on a curated dataset of 1,200 images, preprocessed with extensiveaugmentation to improve robustness under field conditions. Model optimization techniques including quantization and pruning were applied to reduce the sizeto4.3 MB, enabling deployment on Raspberry Pi. Experimental results demonstrated 96.85% classification accuracy with an inference latency of 140ms, outperforming ResNet18 and EfficientNet-B0 baselines while being computationally efficient. The system was integrated with ThingSpeak IoT cloud platformfor remote monitoring and real-time alerts, providing farmers with actionable insights to prevent disease spread. This work demonstrates the potential of edge AIand IoT in achieving cost-effective, scalable, and sustainable precision agriculture solutions.