A Method for Scattering Removal in Three-Dimensional Fluorescence Spectra Based on Physics-Informed Neural Networks

Three-dimensional fluorescence spectroscopy offers high-dimensional information and rapid sensitivity, making it advantageous in fields such as environmental monitoring and food testing. However, scattering effects can hinder analysis accuracy, and traditional de-scattering methods often lead to signal loss and distortion. To overcome these issues, a denoising method based on a physics-informed neural network is proposed. This method includes a physical information reconstruction module that generates theoretical values approximating actual scattering values. Integrated into a GAN framework, this module uses physical quantities to guide the generator in producing physically consistent images, thus preserving material integrity after scattering removal. Tests on a self-built dataset show that this method achieves higher physical consistency compared to traditional methods.