Radar-Based Identification of Insect Species with Ensemble Learning Algorithms Utilizing Multiple Electromagnetic Scattering Parameters

The accurate identification of migratory insect species is pivotal for effective pest forecasting and control strategies. Radar entomology continues to face challenges in insect identification, prompting the exploration of innovative solutions. The precision of conventional insect identification methodologies relying on morphological parameters or radar cross section (RCS) shape was inherently constrained. This study employed ensemble learning algorithms, utilizing multiple electromagnetic scattering parameters of insects as features for species classification, thereby enhancing radar's capability to identify insects. Experiments of measuring insects using two unmanned aerial vehicles (UAVs) were carried out, aiming to establish an electromagnetic scattering database. Data were collected using a multifrequency fully polarimetric entomological radar, capturing echoes from nine major migratory pests in mainland China. Extracting the insect scattering matrix (SM) yielded a total of 22 electromagnetic scattering features categorized into four classes. Three ensemble learning algorithms were employed for classification: random forest (RF), extreme gradient boosting (XGBoost), and stacked generalization (SG). The results demonstrated that the model trained with the XGBoost algorithm consistently exhibited outstanding performance across various frequencies. In the X-band (9.5 GHz, typical operating frequency of entomological radar), the proposed XGBoost algorithm achieved an average identification accuracy of 87.50% for the nine pest species, which is approximately 13% higher than the traditional identification methods based on body size parameters. This study validated the feasibility of insect species identification based on electromagnetic scattering parameters, offering promising prospects for radar entomology to overcome challenges in insect identification.