A Context-Aware IoT and Edge Computing Framework for Wireless Plant Disease Diagnosis Using Compressed MaskRCNN and ResNet-50

Plant disease diagnosis remains a challenging problem in contemporary agriculture, as it relies on physical examination, which is labor-intensive, time-consuming, and often inaccurate at industrial or farmland sites. Timely, precise, and maskable diagnostic activity is increasingly needed to support the improvement of precision agriculture and food security worldwide. To address the need for fast, scalable, and trustworthy diagnostics, this paper introduces a context-aware IoT and Edge Computing solution based on a pruned and knowledge-distilled compressed Mask R-CNN model with a ResNet-50 backbone for real-time plant disease diagnosis. Pruning and knowledge distillation enable the system to be efficiently deployed on low-end edge devices, such as the Raspberry Pi and Jetson Nano. Contextual environmental data, such as temperature, humidity, and moisture, are combined with visual input to improve diagnostic performance under different field conditions. Low-power, low-latency wireless communication is facilitated through MQTT and dynamic frequency transmission based on detection events. The model was trained and validated on a benchmark dataset of diseases in tomato and sugarcane leaves, achieving 91.6% classification accuracy and an F1-score of 90.7%, with only a 2.5% accuracy loss compared to the uncompressed model. The inference latency was reduced to 220 ms in edge devices, with a 38% decrease in power consumption, all for eco-friendly operation. These findings validate the applicability of deep learning models in monitoring plant health in real, low-connectivity scenarios. The proposed solution facilitates early intervention, prevents pesticide misuse, and promotes a data-driven vision for smart agriculture.