Spectral reconstruction of underwater imaging target based on radiative transfer mechanism

Underwater target image reconstruction is constrained by complex light attenuation and scattering effects in water. The raw digital number (DN) values from sensors must undergo precise radiometric calibration and reflectance inversion to reveal the intrinsic spectral characteristics of targets. This paper proposes a spectral reconstruction method based on the Radiative Transfer Mechanism: by integrating the built-in radiometric calibration parameters of hyperspectral cameras, the DN values are converted into radiance (L) according to the pre-calibrated response characteristics of the sensor. Physical modeling of water absorption, scattering, and path radiation effects is incorporated to establish a linear mapping between radiance and reflectance. The Nonnegative Linear Regression Model (NRLM) framework is then used to train a radiative transfer model between DN values and reflectance. By solving the mapping parameters between DN values and reflectance, the true reflectance of target objects is reconstructed based on their DN values. Experiments show that this method can achieve high-precision reflectance reconstruction under different water depths, with excellent performance in indicators such as mean squared error and coefficient of determination. It provides an environmentally adaptive spectral correction solution for underwater target recognition and has application potential in marine resource exploration and underwater imaging.