Designing a TinyML-Enabled early detection system of post harvest maize weevil infestations

Abstract

Post-harvest losses due to maize weevil (Sitophilus zeamais) infestations pose a significant threat to food security and economic stability, particularly in developing countries. This study presents an innovative Tiny Machine Learning (TinyML)-enabled solution for the early detection and realtime monitoring of maize weevil infestations in stored grains. The research integrates acoustic sensing with environmental monitoring to create a comprehensive, non-invasive pest detection system. Over a 101-day period, more than 250,000 audio samples of weevil activity were collected and analyzed, capturing data from both larval and adult stages. Using this extensive dataset, a TinyML model capable of recognizing acoustic signatures indicative of weevil presence was developed and evaluated. The model achieved 98.9% accuracy on the training set and 97.76% on the test set. Through int8 quantization, the model was optimized for deployment on resourceconstrained devices, reducing latency from 157ms to 134ms while maintaining 97.70% accuracy. The system was successfully deployed on Arduino Nano 33 BLE Sense and XIAO ESP32S3 Sense platforms, demonstrating its versatility for various agricultural settings. A web-based dashboard was developed, integrating real-time acoustic detection with environmental monitoring of temperature, humidity, and CO2 levels. This solution offers several advantages over existing methods, including early-stage detection, non-invasive monitoring, and accessibility for small to medium-scale farmers. The system's ability to provide continuous, real-time data enables timely interventions, potentially reducing crop losses and economic impact. This study contributes to the field of smart agriculture by demonstrating the effective application of TinyML and Internet of Things (IoT) technologies in pest management. The developed system not only addresses the immediate need for improved pest detection but also lays the groundwork for future innovations in agricultural technology, promising significant improvements in food security and economic stability for farming communities.