Physically Plausible Spectral Reconstruction in Remote Sensing Using Multispectral Image

Hyperspectral imaging technology offers significant advantages in remote sensing, capturing richer and more detailed spectral information in narrower bands compared to multispectral images. This makes it highly effective for land cover classification, crop monitoring, and environmental monitoring. However, the high cost, low spatial resolution, and complexity of data acquisition limit its widespread application. This study proposes a novel and flexible framework for spectral reconstruction designed to overcome these challenges. Our framework incorporates radiometric calibration, atmospheric correction, classification, spectral unmixing, and thermal infrared super-resolution, all of which can be replaced with the latest high-performance algorithms. Additionally, the physically plausible framework considers actual optical transmission processes, ensuring accurate representation of both reflective and radiative properties. This adaptable approach facilitates cost-effective and scalable acquisition of hyperspectral data. Experimental results demonstrate the accuracy and effectiveness of the proposed method, with reconstructed images showing high consistency with measured data, both in spectral curve shapes and radiance levels, underscoring its significant application potential.