Harnessing mmWave signals and machine learning for noninvasive taxonomic classification of insects.

Monitoring insect biodiversity at sufficiently high resolutions in space and time is crucial to underpin robust and responsive management of terrestrial ecosystems. This study presents a novel approach for taxonomic classification of individual pollinating insects using millimeter-wave (mmWave) signal technology and a hierarchical machine learning (ML) framework. Although ML-based species identification has been proposed for image systems, its application to monitoring and classification of insects remains limited, due to high susceptibility to poor image quality and varying light conditions. The potential of mmWave systems to exploit ML for insect species recognition remains largely unexplored, however. mmWave radar offers access to biomechanical characteristics that are not visible to the human eye or cameras. These characteristics are encoded in the harmonic patterns generated by insect wingbeats and reflected in the radar signal. Here, we systematically explore signal features associated with wing flapping and employ SHapley Additive exPlanations analysis to identify the most discriminative features contributing to classification performance. A hierarchical ML model was developed and achieved an accuracy of 85% in classifying five key pollinator species. Given the growing integration of mmWave systems in communication infrastructure, this method offers a scalable, cost-effective, and contactless solution for high-resolution monitoring of insect biodiversity.