Crop Disease Leaf Image Recognition System Based on CNN and Edge Computing

Abstract

Crop diseases represent a critical threat to global food security. Traditional manual diagnosis approaches are inefficient, while existing cloud-centric AI solutions are plagued by network latency, high data transmission costs, and data privacy risks. This study aims to design and implement an efficient, low-cost edge computing system for real-time crop disease leaf recognition. It investigates the development of a lightweight CNN model tailored for resource-constrained edge devices, its efficient deployment on a Raspberry Pi edge platform, and its advantages over cloud solutions.This paper employs a lightweight Convolutional Neural Network (CNN) based on the MobileNetV2 architecture. The model was trained on the PlantVillage dataset using transfer learning, optimized via post-training quantization, and deployed on a Raspberry Pi edge computing platform. Experimental results demonstrate that the proposed system attains an accuracy exceeding 98%, with a quantized model size of merely 3.2MB. The average inference latency on the Raspberry Pi is less than 500 milliseconds. This edge-centric solution effectively mitigates the inherent limitations of cloud-based paradigms, providing a feasible, practical, and privacy-preserving tool for in-field disease diagnosis, thereby contributing to the advancement of precision agriculture.

Code and data availability

The paper uses the public PlantVillage dataset (cited prior work, not a paper-specific deposit) and describes a MobileNetV2/TFLite model and Raspberry Pi deployment, but provides no author code, model checkpoints, curated subset, or data availability statements with URLs. No qualifying paper-specific public assets are.

No evidence-backed public reproduction asset is currently recorded.