A Wood Anomaly Detection System Based on Electrical Resistivity Tomography and Tiny Machine Learning

This paper aims to investigate the detection of internal voids or decay anomalies in wood using Electrical Resistivity Tomography (ERT) and Tiny Machine Learning (TinyML). It proposes an anomaly detection method suitable for large-scale target detection, aiming to achieve multi-tree detection and automated detection of targets. The ERT tree detection model is simulated using COMSOL and MATLAB software. Based on the obtained boundary voltage data, a machine learning model is generated using NanoEdge AI Studio to simulate the detection of internal anomalies in trees. Finally, control circuits and control programs are designed, and the machine learning model is deployed on STM32G4 series microcontrollers. In the simulation detection using NanoEdge AI, 15 sets of anomalies are detected out of 16 sets of data for the multi-cavity model. The tree anomaly detection method based on Electrical Resistivity Tomography and Tiny Machine Learning can effectively detect anomalies in trees. TinyML conserves embedded system resources during multi-target detection, enabling low power consumption and real-time processing.