Hyperspectral ground object classification method based on multi-kernel machine learning fused with multi-scale features

A multi-scale feature and machine learning technology, applied in machine learning, instruments, computer components, etc., can solve the problems of multiple training parameters, window size and complexity that affect classification accuracy, and achieve high accuracy, reliability, and operation The effect of convenient and fast training

CN111723838APending Publication Date: 2020-09-29中国人民解放军火箭军工程大学
6 Cites 1 Cited by

Patent Information

Authority / Receiving Office
CN · China
Current Assignee / Owner
Publication Date
2020-09-29

Smart Images

  • Figure 1
    Figure 1
  • Figure 2
    Figure 2
  • Figure 3
    Figure 3
Patent Text Reader

Abstract

The invention discloses a hyperspectral ground object classification method based on multi-kernel machine learning fused with multi-scale features. The method comprises the following steps: 1) acquiring a hyperspectral image, performing two-dimensional spatial non-uniform filtering sampling on the hyperspectral image on each waveband by using a non-uniform sampling filtering kernel, stretching andtransforming a two-dimensional matrix obtained by sampling into a one-dimensional vector, and finally performing dimension reduction on the two-dimensional matrix by using PCA to extract a principalcomponent waveband to obtain features of the hyperspectral image; and 2) extracting multi-scale features of the target; carrying out organic fusion on multi-scale features and multi-scale kernels of the target in a multi-kernel learning framework formula; weighted mapping, obtaining a fusion kernel matrix, then obtaining an optimal multi-scale kernel classifier by training a decision function, finally, using the optimized multi-scale kernel classifier for classifying images, and completing hyperspectral ground object classification based on multi-core machine learning fusion multi-scale features, and hyperspectral ground object classification can be conducted in a high-precision and high-real-time mode.
Need to check novelty before this filing date? Find Prior Art

Description

technical field

[0001] The invention relates to a method for classifying hyperspectral objects, in particular to a method for classifying hyperspectral objects based on multi-core machine learning and fusion of multi-scale features. Background technique

[0002] Hyperspectral remote sensing, as a multi-dimensional information acquisition technology that combines remote sensing imaging and spectral subdivision, greatly enriches the information of acquired image targets. For the most concerned classification technology in hyperspectral image applications, the map integration, high-dimensional characteristics of hyperspectral data, and how to preprocess, unmix, effective feature extraction, and target recognition of hyperspectral images have always been urgently needed. problem. As the resolution of hyperspectral imagers continues to increase, hyperspectral images are widely used in resource surveys, crop yield estimates, disaster forecasts, mineral deposit exploration, topogr...

Examples

Embodiment Construction

[0036] The present invention is described in further detail below in conjunction with accompanying drawing:

[0037] refer to Figure 1 to Figure 4 , the multi-core machine learning fusion multi-scale features of the present invention hyperspectral object classification method comprises the following steps:

[0038] 1) Imitating the non-uniform sampling of visual information of the biological retina and the selective biological visual attention mechanism, the present invention considers the construction of a multi-scale space based on non-uniform sampling, in which uniform sampling is performed within the fovea, and the sampling rate is the highest is 1, that is, the original resolution of the image, and the sampling rate is decreased by twice the distance from the periphery to the center point, so the input information can be compressed to reduce the difficulty of storage and processing.

[0039] 2) Extract the features of the hyperspectral image in different scale spaces, s...