Abstract: Using Global Navigation Satellite System-Scattering (GNSS-S) signals as external radiation sources for sea surface moving target detection offers benefits such as extensive coverage, long detection times, and abundant signal sources. However, challenges like strong sea clutter, weak target echoes, and low signal-to-noise ratios make detection difficult. To address this, we propose an autoencoder-based GNSS-S multi-frequency, multi-polarization radar method. The method preprocesses dual-frequency, dual-polarization echoes through partitioning, expansion, and filtering to extract moving target signals from clutter. The signals are processed regionally, fused via an autoencoder, and then reconstructed for CFAR target detection. Experimental results demonstrate effective performance in sea surface target detection, improving detection probability and range accuracy, and laying a foundation for multi-satellite target tracking.