An Efficient 3D Network for Enhanced Accuracy in Diffractive based Spectral Image Reconstruction

This document introduces an efficient 3D network for enhanced accuracy in diffractive based spectral image reconstruction.The input dataset consists of spectral images captured through diffractive optical imaging spectroscopy, including both original and diffracted images.Pre-processing steps, such as normalization and resizing, are applied to optimize the image data for neural network performance.The proposed method utilizes a 3D CNN with a U-Net architecture to improve both precision and speed of image reconstruction.The model is trained and tested on separate datasets to learn the mapping between diffracted and reconstructed images.Performance is evaluated using metrics such as Mean Peak Signalto-Noise Ratio (MPSNR) and Mean Structural Similarity Index (MSSIM), which measure image quality and visual similarity.Additionally, the algorithm is optimized to achieve real-time spectral image reconstruction, offering a significant improvement in both accuracy and computational efficiency for practical applications.