DAEC-UCT: an improved UCTransNet for SAR road extraction

Synthetic Aperture Radar (SAR) images are widely used in geographic information extraction, post-disaster assessment, and intelligent transportation. Road extraction from SAR images plays a vital role in tasks such as intelligent navigation, emergency rescue, and urban planning. However, traditional SAR road extraction algorithms face challenges in resolution limitation, complex road structures, and background interference, making accurate identification challenging. To address these challenges, this paper, for the first time, applies the UCTransNet network to SAR road extraction and proposes an improved network: Dual Attention and Enhanced Convolution UCTransNet (DAEC-UCT). The proposed model integrates a DEConv detail-enhanced convolution module, a Refined Efficient channel and Spatial Attention mechanism (RESA), and a Wavelet-based Frequency-domain Attention module (WFA) to enhance structural integrity and road discrimination while suppressing background interference. Compared with mainstream image segmentation algorithms, the proposed model achieves superior performance on SAR road datasets with varying resolutions in terms of F1-score, IoU, Precision, and Recall, demonstrating its effectiveness in the SAR road extraction.