Extraction method for identification characteristic of high spectrum remote sensing data

Abstract

The invention provides an extraction method for identification characteristic of high spectrum remote sensing data, which comprises the following steps of: reading and inputting high spectrum remote sensing image data into a computer, generating a vector with each data point in the high spectrum remote sensing image according to the wave band of the data point, forming a matrix with the whole high spectrum remote sensing image to be used as a training sample set, selecting partial data points from the training sample set to label known ground object categories and form a sample category label, at the precondition of the known ground object categories of the partial data points, constructing a similar diagram and a dissimilar diagram with the training sample set to measure the similarity and the diversity of the data points, respectively counting a weighting matrix according to the constructed similar diagram and dissimilar diagram, counting a local similar structure matrix and a local dissimilar structure matrix, counting a projection matrix with a goal optimizing function, and projecting the high spectrum remote sensing data to a low order embed space with the projection matrix to extract the identification characteristic of high spectrum data. The invention effectively solves the extraction problem of an intrinsic manifold structure and an identification characteristic in the high spectrum data.

Description

technical field [0001] The invention relates to the technical field of hyperspectral data processing methods and applications, in particular to a hyperspectral remote sensing data identification feature extraction method. Background technique [0002] Hyperspectral remote sensing is one of the major technological breakthroughs that humans have made in earth observation in the last two decades of the 20th century. In the imaging process, it uses hundreds of continuous narrow spectral bands to describe a pixel with extremely high spectral resolution, and generates a complete and continuous spectral curve for each pixel while providing an image of each band interval. Hyperspectral remote sensing images contain rich triple information of space, radiation and spectrum, and have wide application and development space in related fields. The characteristics of hyperspectral data are as follows: a) Unification of spectra: while acquiring hundreds of spectral images, the continuous s...

AI Technical Summary

Problems solved by technology

However, the LDA algorithm also has defects, such as only extracting features that are less than the number of labeled categories, without considering the difference in category variance, and the small sample problem will make the algorithm invalid

The supervised manifold learning method needs to label all the data in the sample data to provide the data to be tested for learning; however, in many machine learning methods, the data that can be labeled is relatively limited, especially hyperspectral remote sensing data, Due to the large amount of data, high redundancy, and low signal-to-noise ratio, a hyperspectral remote sensing image can be marked with very limited ground object data, and it is impossible to achieve all the labeling of sample data to provide learning

Images