LABLF: a multispectral and SAR image fusion method based on least squares-optimized adaptive box-guided and Laplacian-Gaussian filtering

The fusion of multispectral (MS) and synthetic aperture radar (SAR) images can provide complementary multidimensional information, thereby improving land cover classification accuracy. However, an imbalance between spatial detail enhancement and spectral information preservation often occurs during the fusion process. To address these issues, an MS and SAR image fusion method is proposed, which integrates a least squares-optimized adaptive box-guided (LAB) and Laplacian-Gaussian (LG) filtering, denoted as LABLF, for multi-scale fusion. First, channel enhancement and Laplacian energy injection are performed on both MS and SAR images to improve their visual quality. Then, the images are decomposed using LAB combined with LG filtering. Finally, advanced fusion rules are applied to achieve hierarchical image feature fusion. The final results demonstrate that the proposed method achieves superior visual quality compared to other approaches and achieves the best overall performance in nine evaluation indicators. In the validation experiments for the downstream land cover classification task, the proposed method demonstrated average improvements of 2.838% in overall accuracy and 0.041 in the Kappa coefficient compared to the original MS images. The source code is available at https://github.com/RSIDEA-ECUT/LABLF.