Spectral Micro-Unit-Based Insect-Bird Classification Using Polarimetric Weather Radar

Insect-bird classification using weather radar is a key component of large-scale aeroecological monitoring and supports applications such as bird-strike risk warning and pest surveillance. However, conventional radar-based products describe only bulk properties of radar resolution cells and fail to capture fine-scale internal structures, limiting accurate classification when insects and birds coexist within the same cell. To address this limitation, this study proposes an insect-bird classification method based on spectral micro-unit features derived from weather radar spectral signals. Spectral micro-units associated with biological scatterers are extracted through spectral peak matching between horizontal and vertical polarization channels. Based on differences in morphology and flight behavior between insects and birds, a spectral micro-unit feature system is constructed to characterize intensity, structural dispersion, and symmetry. An unsupervised feature-space clustering approach is then employed to achieve insect-bird classification without manual labeling. Experimental results demonstrate that the proposed method achieves a clustering silhouette coefficient (SC) of 0.703, effectively separating insects and birds and enabling reliable identification of mixed insect-bird targets within individual radar resolution cells. The proposed method establishes a spectral-level analytical framework for insect-bird classification and provides a reference for future fine-grained classification at higher taxonomic levels.