The paper presents a novel approach to land cover classification (LCC) by fusing synthetic aperture radar (SAR) and optical images. The authors propose a dual-input model that utilizes image-level fusion and feature-level fusion to bridge the gaps between SAR and optical images. The image-level fusion is achieved using principal component analysis (PCA) to reduce data dimensionality and retain important features. Feature-level fusion involves fusing shallow feature maps rich in geometric features and incorporating a channel attention module to highlight relevant information. The proposed method is validated on two public datasets, the WHU-OPT-SAR dataset and the DDHRNet dataset, showing significant improvements in mIoU metrics for various land cover classes. The results demonstrate that the fusion of SAR and optical data enhances the accuracy and completeness of LCC, particularly in distinguishing features such as farmland, city, village, water, and road. The study also highlights the effectiveness of the channel attention module in improving feature extraction and reducing false positives. Overall, the proposed multimodal fusion strategy is shown to be effective and robust for LCC tasks.The paper presents a novel approach to land cover classification (LCC) by fusing synthetic aperture radar (SAR) and optical images. The authors propose a dual-input model that utilizes image-level fusion and feature-level fusion to bridge the gaps between SAR and optical images. The image-level fusion is achieved using principal component analysis (PCA) to reduce data dimensionality and retain important features. Feature-level fusion involves fusing shallow feature maps rich in geometric features and incorporating a channel attention module to highlight relevant information. The proposed method is validated on two public datasets, the WHU-OPT-SAR dataset and the DDHRNet dataset, showing significant improvements in mIoU metrics for various land cover classes. The results demonstrate that the fusion of SAR and optical data enhances the accuracy and completeness of LCC, particularly in distinguishing features such as farmland, city, village, water, and road. The study also highlights the effectiveness of the channel attention module in improving feature extraction and reducing false positives. Overall, the proposed multimodal fusion strategy is shown to be effective and robust for LCC tasks.