Abstract: Conventional methods for sea ice concentration (SIC) inversion are constrained by specific frequency bands and the inherent limitations of single algorithms. These constraints make it difficult to fully utilize the full-band information captured through passive microwave remote sensing. Especially in the summer melt season, the inversion of SICs in thin-ice regions faces significant challenges, including large errors and low accuracy. These technical bottlenecks severely restrict the precise monitoring of the Arctic Northeast Passage. Additionally, although China's Fengyun-series satellites enable multi-band observation, research based on these resources remains relatively scarce. To address these challenges, this study develops an SIC inversion model, i.e., SIC-VCDNet, based on data from the microwave radiation imager (MWRI) onboard the FY-3D satellite. This model employs a dual-branch network that integrates VMamba with an attention mechanism, enabling deep mining of full-band features. First, a high-precision product dataset is constructed through the weighted fusion of traditional algorithms (ASI, BT, and NT2) and the incorporation of ship-based in-situ measurements. Subsequently, the SIC-VCDNet model with a dual-branch network architecture is built by combining VMamba with the convolutional block attention module (CBAM), thus achieving the effective fusion of global contextual dependencies and local fine-grained features. The model was used to identify both local details, such as sea ice edges and melt ponds, and the global distribution of ice conditions based on full-band data covering 10~89 GHz. The results indicate that the proposed model significantly improved the inversion accuracy compared to other methods, with R² reaching 0.992 and the root mean square error (RMSE) reduced to 2.543. These results demonstrate that the developed model delivers excellent performance in the summer melt season. Analysis of ice conditions along the Arctic Northeast Passage based on inversion results reveals that the area to the north of the New Siberian Islands emerges as a navigational bottleneck, with an annual navigable window of only 56 d. In contrast, areas such as the Kara Strait allow for year-round navigation. These findings provide a crucial spatiotemporal reference for the scientific planning and safe navigation of Arctic shipping routes. This study introduces a novel approach for high-precision SIC inversion while also providing crucial technical support for the in-depth application of data from China's self-developed satellites.