Spectral Image Reconstruction Using a Pre-Calibrated Sensing Matrix
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional spectral image reconstruction methods are time-consuming and result in low-resolution images due to the use of complex, bulky, and expensive precision optical components, and the iterative optimization algorithms employed are slow, leading to mosaic-like reconstructions.
Innovation Solution
A reconstruction method utilizing a pre-calibrated sensing matrix and convolutional neural networks for noise reduction, including linear transformations and noise reduction processes through U-net, auto-encoder, and residual neural network structures, to quickly generate high-resolution spectral images without mosaicking.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional iterative optimization algorithms are used for spectral image reconstruction, then spectral information can be recovered, but the reconstruction process is slow and time-consuming
Solution Approach 1:
The patent replaces the traditional iterative optimization algorithm (mechanical computation process) with a neural network model that performs spectral reconstruction through learned patterns. The neural network is trained offline using iterative optimization, but during actual spectral imaging, it performs rapid inference without iterative computation, thus substituting the time-consuming mechanical iteration process with a faster neural network evaluation process.
Solution Approach 2:
The patent performs the computationally intensive iterative optimization process in advance during the offline training phase to pre-train the neural network model. This preliminary action allows the model to learn the optimal reconstruction patterns beforehand, so that during actual spectral imaging applications, the reconstruction can be performed rapidly without repeating the slow iterative process.
2Measurement precision
If traditional iterative optimization algorithms reconstruct spectrum of each point one by one, then spectral information can be obtained, but spatial resolution is reduced and mosaic characteristics appear
Solution Approach 1:
The patent merges the reconstruction process across multiple spatial points by using a neural network that processes the entire measurement image simultaneously. Instead of treating each pixel independently as in traditional point-by-point iterative methods, the neural network learns spatial correlations and jointly optimizes the reconstruction across the whole image, thereby preserving spatial resolution and eliminating mosaic artifacts.
Solution Approach 2:
The neural network model serves multiple functions: it performs spectral reconstruction, preserves spatial resolution, and eliminates mosaic artifacts all within a single unified framework. The model is trained to simultaneously optimize for spectral accuracy and spatial fidelity, making it a universal solution that addresses multiple limitations of traditional point-by-point iterative reconstruction methods.
Data Source
AI summary
A reconstruction method for spectral image includes: obtaining a measurement image of an imaging object, and reconstructing and obtaining a spectral image of the imaging object according to the measurement image and a pre-calibrated sensing matrix. The spectral image includes spectral information at different position points of the imaging object. Through the method, defects of the time-consuming spectral image reconstruction method and the low resolution of the obtained spectral image in the related art can be overcame and the spectral image of the target imaging object can be obtained by reconstructing quickly and the obtained spectral image has high spatial resolution and no mosaic.