Next-Frequency Autoregressive Modeling for Remote Sensing Change Detection

Abstract

Remote sensing change detection identifies land-cover changes from bi-temporal images. Most existing methods follow a one-shot dense prediction paradigm, directly regressing a change mask from fused features, but overlook the intrinsic frequency characteristics of change patterns. We propose Freq-RemoteVAR, a frequency autoregressive framework that reformulates change detection as structured generation in the frequency domain. Instead of predicting the change mask in one step, we introduce next-frequency prediction, where change information is progressively generated from coarse to fine. We design a frequency-aware mask tokenization strategy that decomposes change supervision into multi-frequency token targets via Fourier transformation and quantization. Based on this representation, we develop a Frequency VAR Transformer, which performs causal autoregressive modeling over frequency tokens. Starting from learned mask queries, it progressively predicts frequency-level tokens conditioned on previously generated tokens and bi-temporal image features, capturing long-range dependencies across frequency scales. We further introduce Scale-Aligned RoPE Cross Attention (SRCA), which aligns frequency-domain mask queries with spatial-domain bi-temporal features under a unified coordinate system, enhancing spatial-frequency consistency. To mitigate unreliable signals, a Change-quality Control module adaptively modulates generation through dynamic normalization, attention biasing, and spatial offset adjustment, suppressing pseudo-change responses and improving robustness. Experiments on CDD, GZ-CD, and LEVIR-CD show that Freq-RemoteVAR consistently outperforms existing methods, especially in challenging scenarios with complex appearance variations and noisy disturbances.