FS-SWTnet: a swin transformer-based multi-domain feature fusion network for radar detection of LSS targets

Abstract Current methods for detecting low, slow, and small targets typically rely on a single type of signal feature, such as time-domain, frequency-domain, or Doppler characteristics. This narrow reliance can limit the completeness of target representation. To address this limitation, this study proposes a multi-feature fusion model, termed FS-SWTnet, which integrates both time-domain and frequency-domain information. The model adopts a cross-fusion strategy that allows the two feature domains to complement each other, enabling richer representations at different semantic levels. To evaluate the influence of class granularity on recognition performance, experiments are conducted under multiple grouping scenarios. FS-SWTnet achieves consistently high classification accuracy, including 99.44% for the four-drone subset, 99.24% for the four-class set (two drones, one bird, and one pedestrian), 97.79% for the six-drone subset, and 96.11% for the nine-class set (six drones, two bird species, and pedestrians).