A Data Enhancement Method for Rock Hyperspectral Image

Abstract

The invention belongs to the field of geological exploration, and in particular relates to a rock hyperspectral image data enhancement method. The method comprises the following steps: step (1) collecting the hyperspectral image in which the pixel spectrum verified by the ground rock analysis test is a sample for data enhancement; Step (2) calculate the spectral reflectance x of each mineral endmember; step (3) calculate the single scattering albedo of each mineral endmember; step (4) calculate the single scattering albedo of mixed minerals; step ( 5) Convert the mixed mineral spectrum constructed in step (4) into the spectral reflectance in the image; step (6) combine the mixed mineral spectral reflectance from different angles to produce a data-enhanced rock hyperspectral image. The present invention has done a lot of experimental research on the spectra of granite, carbonatite and other rocks, and constructed a spectral sample data set that can be used for deep learning.

Description

technical field [0001] The invention belongs to the field of geological exploration, in particular to a rock hyperspectral image data enhancement method. Background technique [0002] The accurate inversion of surface mineral content by hyperspectral technology plays an important role in geological survey and mineral exploration. Although many models and machine learning-based methods have been applied to hyperspectral mineral identification and content assessment, due to the influence of nonlinear mixing, sensor noise, atmosphere, terrain and other factors, traditional inversion models inevitably bring error. With the wide application of deep learning, deep neural network has been widely used in mineral spectrum analysis as a nonlinear analysis method. Spectral sample data with known components, too little sample data can easily make the network fall into the phenomenon of local optimal solution and overfitting. [0003] Therefore, it is necessary to establish a fast and...

AI Technical Summary

Problems solved by technology

Although many methods based on model and machine learning have been applied to hyperspectral mineral identification and content assessment, due to the influence of nonlinear mixing, sensor noise, atmosphere, terrain and other factors, the traditional inversion model inevitably brings error

With the wide application of deep learning, deep neural network as a nonlinear analysis method has been widely used in mineral spectral analysis, but the main problem in the current application is that there is less data available for training, and deep neural network requires a large number of existing methods. Spectral sample data with known components, too little sample data will easily make the network fall into local optimal solution and over-fitting phenomenon

Images