Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

Hyperspectral remote sensing data feature extraction method based on convolution neural network

A convolutional neural network and hyperspectral remote sensing technology, applied in the field of remote sensing data feature extraction, can solve problems such as unfavorable image processing and feature data without deep structure

Inactive Publication Date: 2015-07-01
HARBIN INST OF TECH
View PDF5 Cites 23 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0003] The purpose of the present invention is to solve the problem that the existing method is used for feature extraction of hyperspectral remote sensing data, and the obtained feature data does not have a deep structure, which is not conducive to subsequent image processing, and provides a hyperspectral remote sensing based on convolutional neural network Data feature extraction method

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Hyperspectral remote sensing data feature extraction method based on convolution neural network

Examples

Experimental program
Comparison scheme
Effect test

specific Embodiment approach 1

[0015] Specific implementation mode one: the following combination figure 1 Illustrate this embodiment, the hyperspectral remote sensing data feature extraction method based on convolutional neural network described in this embodiment, it comprises the following steps:

[0016] Step 1: Perform bad band removal, data deformation reorganization and data preprocessing on the hyperspectral raw data to obtain standardized input data;

[0017] Step 2: The standardized input data is used as the input of the convolutional neural network convolution layer, and the standardized input data is convoluted through n trainable filters and offsets to obtain n different feature maps, where n is Positive integer; each neuron in each feature map is connected to a local receptive field of standardized input data for extracting corresponding local features, and the corresponding local features obtained by all neurons in each feature map are integrated Finally, a global information is obtained, an...

specific Embodiment approach 2

[0025] Embodiment 2: This embodiment further explains Embodiment 1. The removal of bad bands specifically includes: deleting bands in the original hyperspectral data whose bands are all zero values ​​or whose noise exceeds a preset noise threshold, and obtain a three-dimensional matrix. No bad bands to enter data.

[0026] In the process of hyperspectral imaging, there are some bands that are all zero or noisy due to water absorption or other reasons. In the bad band removal part, these bands need to be deleted to avoid singularity in subsequent data processing and meet the needs of subsequent algorithms.

specific Embodiment approach 3

[0027] Specific implementation mode three: this implementation mode further explains implementation mode two, and the data deformation and reorganization is specifically: extracting the pixel of each pixel point from the input data without bad bands in order from left to right and from top to bottom Vector, each pixel vector is used as a row vector, and all the row vectors are arranged downwards in order to form a two-dimensional matrix input data. The number of rows of the two-dimensional matrix input data is the number of pixels of the input data without bad bands. The number of columns is the number of bands each pixel contains.

[0028] In this embodiment, the original hyperspectral data after bad band removal is reorganized to meet the data input requirements of the convolutional neural network. The original hyperspectral data can be regarded as a three-dimensional matrix, and the vector formed by different band positions at the same spatial position is called a pixel vec...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The invention discloses a hyperspectral remote sensing data feature extraction method based on a convolution neural network, belongs to the technical field of remote sensing data feature extraction and aims to solve the problems that when the hyperspectral remote sensing data features are extracted by the existing method, the acquired feature data are not provided with deep structures and the post image processing is not benefited. The method includes firstly, performing bad band removal, data transformation and reconstruction and data pre-processing on the hyperspectral original data, and acquiring standardized input data; performing convolution on the standardized input data by different filters through localized receptive fields; performing sub sampling on the convolution results; stacking the convolution layer and the sub sampling layer, acquiring standardized input data state response, and implementing the hyperspectral remote sensing data feature extraction. The method is applied to the hyperspectral remote sensing data feature extraction.

Description

technical field [0001] The invention relates to a hyperspectral remote sensing data feature extraction method based on a convolutional neural network, and belongs to the technical field of remote sensing data feature extraction. Background technique [0002] With the development of imaging technology and spectral technology, hyperspectral remote sensing technology has become a cutting-edge technology in the field of remote sensing. The improvement of spectral resolution has solved problems that cannot be solved by other remote sensing technologies, but the huge amount of data has increased the difficulty of feature extraction technology. In order to solve this problem, some linear dimensionality reduction methods have been proposed, such as principal component analysis, independent component analysis, factor analysis, etc. For hyperspectral complex spectral information, nonlinear methods can also describe data characteristics well. Popular learning is a commonly used nonline...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
Patent Type & Authority Applications(China)
IPC IPC(8): G06K9/46G06N3/02
Inventor 陈雨时姜含露陈曦
Owner HARBIN INST OF TECH
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products