A Dimensionality Reduction Method for Spectrum Reconstruction from a Single RGB Image

Abstract

The invention discloses a dimensionality reduction method for reconstructing a spectrum of a single RGB image, comprising the following steps: step (1), selecting a sample set, setting a training set and a test set; step (2), setting a regularization parameter set and The number of hidden layers is set; step (3), set the number m of extracted feature vectors, for the training set, select the regularization parameter value and hidden layer value arbitrarily respectively, use the improved nonlinear principal component analysis method to train, and obtain n groups of extracted The model of the feature vector; step (4), according to the minimum average error value, select the regularization parameter value p and the hidden layer value o; step (5), obtain the target feature vector. The few-dimensional feature vector obtained by the method of the invention can reduce the ill-conditioned problem of the reconstruction equation in reconstructing the spectrum, and improves the precision and stability of the reconstruction.

Description

technical field [0001] The invention belongs to the technical field of spectral imaging, and in particular relates to a dimensionality reduction method for reconstructing a spectrum from a single RGB image. Background technique [0002] Spectral images composed of reflectance have been widely used in color and material analysis in cultural heritage protection, disease analysis and judgment in medical science, and identification of geological landforms in earth science and remote sensing. At present, the acquisition equipment of spectral images are mainly spectral cameras, and equipment composed of broadband filters or narrow-band light sources and ordinary cameras. Such spectral imaging equipment is either expensive (such as a spectral camera), or complicated in process and not easy to carry, so it is not easy to be popularized and used in fields that require low price and easy portability. In order to avoid the aforementioned problems, in the prior art, a digital camera (l...

AI Technical Summary

Problems solved by technology

The dimensionality reduction method in the prior art can easily lead to overfitting, thus affecting the reconstruction accuracy of the reflectance spectrum of unknown samples, which is contrary to practical applications

Images