Optimized Deep Learning Models For Edge Computing: A Comparative Study on Raspberry PI4 For Real-Time Plant Disease Detection

In the pursuit of precision agriculture, the integration of artificial intelligence (AI) for real-time plant disease detection holds a significant promise. This study investigates the application of edge computing on a Raspberry Pi 4 platform to keep the performance of three optimized deep learning models: GoogLeNet, ResNet and XceptionNet, in detecting plant diseases using the new plant diseases dataset. The goal seeks to bridge the gap between theoretical advancements in deep learning and practical applications in precision agriculture, shedding light on the trade-offs between model complexity, computational demands, and diagnostic accuracy. Additionally, we provide a lightweight web application that interfaces with the trained models, exemplifying how edge computing can democratize access to AI tools, making them accessible to farmers and researchers. Our findings show that converting the models to TensorFlow Lite significantly reduces model size and inference time while maintaining similar performance, crucial for applications on resource-limited devices like the Raspberry Pi 4.