Eureka AIR delivers breakthrough ideas for toughest innovation challenges, trusted by R&D personnel around the world.

Hyperspectral band selection method based on separable convolution and hard threshold function

A band selection and hard threshold technology, applied in image processing and agriculture, can solve the problems of slow speed, poor band selection performance, low classification accuracy, etc., to improve classification accuracy, speed up band selection and classification speed, and fast classification speed. Effect

Active Publication Date: 2019-02-15
XIDIAN UNIV
View PDF9 Cites 17 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

It is used to solve the technical problems of poor band selection performance, low classification accuracy and slow speed in the existing band selection-based hyperspectral image classification 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 band selection method based on separable convolution and hard threshold function
  • Hyperspectral band selection method based on separable convolution and hard threshold function
  • Hyperspectral band selection method based on separable convolution and hard threshold function

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0036] Because hyperspectral data records continuous spectral information and has high spectral resolution, it has a wide range of applications in military surveys, ecological construction, land use, global environment, natural disasters, and other fields. However, due to the high resolution of the hyperspectral image spectrum, it has a high spectral dimension and a large amount of data, which brings many problems to the processing of hyperspectral images, such as increasing information redundancy and data storage. space, and prolong the data processing time. In addition, hyperspectral images contain noise or invalid bands caused by interference from other factors during the radiation process. Crucially, when the number of image samples is small and the amount of processed data is large, It is easy to produce the phenomenon of "curse of dimensionality". Therefore, it is extremely necessary to reduce the number of bands in hyperspectral images.

[0037] There are existing hype...

Embodiment 2

[0059] The hyperspectral band selection method based on separable convolution and hard threshold function is the same as embodiment 1, and the construction band selection layer described in step (4a) is used as the first layer of the convolutional neural network based on band selection, including the following steps :

[0060] (4a1) The convolutional neural network based on band selection has a total of ten layers. Among them, the band selection layer based on separable convolution is the first layer, and the space-spectrum joint information extraction layer based on multi-scale convolution is the second layer. Based on three A classifier with one convolutional layer, two pooling layers, two fully connected layers and one softmax classification layer is used as the follow-up network.

[0061] The number of network layers above can also be modified, but it is worth noting that if the number of network layers is too small, the nonlinear classification ability of the network will...

Embodiment 3

[0066] The hyperspectral band selection method based on separable convolution and hard threshold function is the same as embodiment 1-2, and the band selection layer described in step (4b) utilizes separable convolution and hard threshold function to select bands, including the following steps:

[0067] (4b1) Arrange the weight matrix of the band selection layer in descending order to obtain a one-dimensional vector S with a size of 1×200, determine the number N of selected bands, and use the elements at the numerical positions of the bands in the sorted vector as thresholds, That is, the element at the Nth position in the vector S is taken, and a hard threshold function based on the threshold is established.

[0068] (4b2) Enter the hard threshold function obtained by inputting each element in the weight matrix of the band selection layer. Specifically, for the weight value greater than or equal to the threshold value, the output retains the original value, otherwise it is set...

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 provides a hyperspectral band selection method based on separable convolution and hard threshold function, which solves the problems of poor band selection performance, poor classification accuracy and long time consumption of hyperspectral image. The method comprises the following steps of inputting a hyperspectral image; generating training and test samples; constructing a convolution neural network including band selection layer, spatial-spectral joint information and classifier, obtaining the optimized convolution neural network by training; classifying hyperspectral images by using the trained convolution neural network based on band selection. The method of the invention constructs the convolution neural network based on band selection, builds a band selection layer based on separable convolution and hard threshold function in the network structure, realizes the integration of band selection and classification, and overcomes the problems of low classification precision and slow speed caused by the separation of the band selection and the hard threshold function. The learning ability of network features is used to improve the performance of band selection and classification accuracy. The method is widely used in military, civil, agriculture, ecology and other fields.

Description

technical field [0001] The invention belongs to the technical field of image processing, and relates to a hyperspectral image classification, specifically a hyperspectral band selection method based on separable convolution and hard threshold functions, which is used in the fields of agriculture, surveying and mapping, archaeology, environment and disaster monitoring, etc. . Background technique [0002] With the development of remote sensing technology and imaging technology, the application fields of hyperspectral remote sensing technology are becoming more and more extensive. Hyperspectral data can be regarded as a three-dimensional data cube, which adds one-dimensional spectral information in addition to ordinary two-dimensional image data. Hyperspectral remote sensing images combine rich spatial domain information and spectral domain information, and have the characteristics of "integration of map and spectrum", which provides higher discrimination for accurate identif...

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): G06K9/00G06K9/62
CPCG06V20/194G06V20/13G06F18/2413
Inventor 冯婕陈建通冯雪亮焦李成张向荣王蓉芳刘若辰
Owner XIDIAN UNIV
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Eureka Blog
Learn More
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic, Popular Technical Reports.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com