Pollinating insects are important for agriculture and ecological flourishing, but they are difficult to monitor, as identification is tricky, labor-intensive, and typically requires killing some insects. Adam Narbudowicz and colleagues use machine learning to identify insects by the changes in their radar reflection, caused by the flapping of their wings. The machine learning model extracted more than 70 harmonic, spectral, and temporal features from the Doppler radar signatures. To train the model, insects were captured on the campus of Trinity College Dublin and placed individually in small cylindrical plastic containers which were placed on top of a millimeter-wave antenna. After their radar signatures were recorded, the insects were released. After extracting relevant micro-Doppler features from the data, the model was trained. Key distinguishing features included how quickly an insect’s wing movements change, as well as its fundamental wing beat frequencies, among others. The model was able to distinguish between bees and wasps with an accuracy of 96% and classify five insects to the species level with an accuracy of 85%. According to the authors, millimeter-wave radar reflections from insect wingbeats, perhaps measured via a fly-through device, could be used for cheap, non-lethal monitoring of insect biodiversity.
PNAS Nexus
Harnessing mmWave signals and machine learning for noninvasive taxonomic classification of insects
28-Apr-2026