Fusion of Hyperspectral and Multispectral Images for Land Use Segmentation

Land use feature detection is possible through high spatial and spectral resolution images. The spectral image resolution of hyperspectral sensors is mostly high, but the spatial resolution is low. The multispectral sensor produces good spatial resolution images with poor spectral resolution. In recent years, numerous techniques have been proposed for the improvement of spatial and spectral resolution in high-quality images acquired from the same scene by combining the acquired data. For this purpose, we fused hyperspectral and multispectral images, removed distortions, and improved the quality of the images by Gram–Schmidt, one of the pan-sharpening methods. Our selected study area is in Beijing, China. The fused image generated from the same area is utilized to segment land use through the U-Net model. The model we used for this study is trained and tested fused, multispectral, and hyperspectral images, and results showed that overall accuracy is 86.88 for hyperspectral imagery, 93.28 for multispectral imagery, and 97.59 for fused imagery. The results signify the employed approach has improved both image enhancement and fusion of images. Additionally, this technique can generate an imaging output characterized by high spectral data quality and high spatial resolution.