CLFDA: Continuous Low-Frequency Decomposition Architecture for Fine-Grained Land Cover Classification

Fine-grained land cover classification from high-resolution remote sensing imagery plays a vital role in urban and environmental monitoring. While the existing spatial-domain-based approaches achieve notable progress, their performance in complex scenarios remains constrained by insufficient modeling of characteristics. This letter proposes the continuous low-frequency decomposition architecture (CLFDA) to address insufficient cross-domain modeling of multiscale frequency characteristics in current methods. The architecture introduces frequency-domain features through continuous low-frequency decomposition, where each frequency decomposition and enhancement (FDE) module employ discrete wavelet transform (DWT) to separate spatial and low-frequency features into low-frequency and high-frequency subbands. Low-frequency features feedback into the encoder for global context, while high-frequency features are routed to the decoder via attention mechanisms for detail refinement, enabling bidirectional spatial–frequency fusion. By integrating convolutional neural networks (CNNs), vision transformer, and mamba backbones, our CLFDA achieves 2.0% and 3.46% average mIoU improvements on the GID-15 and the FUSU datasets, respectively. These consistent performance gains across heterogeneous backbones demonstrate the effectiveness and generalizability of our CLFDA in modeling frequency-domain features. The code is at https://github.com/GeoRSAI/CLFDA