Lightweight Pest and Soil Moisture Detection with MobileViT and IoT-Ethereum Hybrid Blockchain

Abstract

Pest detection and soil moisture estimation models with little computation overhead are needed in resource-efficient pesticide monitoring of agricultural fields. This paper introduces a lightweight MobileViT-based system that is combined with IoT sensors and a hybrid Ethereum blockchain platform to offer secure and real-time pest and soil monitoring. MobileViT is a hybrid architecture that uses convolutional networks and transformer-based global features, which allow competition with detection accuracy and low computing needs. The IoT sensors are used to monitor soil moisture, temperature, and humidity to aid in making irrigation decisions with the key events being safely stored on the Ethereum blockchain to trace the events irrevocably. The system is executed in Python using PyTorch, OpenCV, and Web3.py and runs on edge devices and is fast in inference with low latency. The IP102 dataset includes the evaluation which proves that the model has high detection performance with accuracy 92.8%, precision 91.5%, recall 90.7%, F1-score 91.1%.