Hyperspectral image classification method based on auto-encoder and 3D deep residual network

A hyperspectral image and autoencoder technology, which is applied in the field of hyperspectral image classification based on autoencoder and 3D deep residual network, and can solve problems such as classification of difficult remote sensing hyperspectral images.

Active Publication Date: 2021-01-15
ANHUI UNIVERSITY
View PDF3 Cites 17 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0006] The purpose of the present invention is to solve the defect that it is difficult to classify remote sensing hyperspectral images in the

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 image classification method based on auto-encoder and 3D deep residual network
  • Hyperspectral image classification method based on auto-encoder and 3D deep residual network
  • Hyperspectral image classification method based on auto-encoder and 3D deep residual network

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0174] Embodiment 1: In order to verify the impact of different parameter settings on the classification accuracy in the method proposed by the present invention, several factors that affect the classification effect of the model in the SAE-3DDRN method will be analyzed, namely, the dimensionality reduction d of the self-encoding network and the input sample size Size w, residual network structure, learning rate size lr. For the convenience of description, this section will randomly select 10% of the training samples from each type of ground object in the Indian Pines dataset, and the rest will be used as test samples, and the average of the classification accuracy of 10 experiments will be used as the experimental result.

[0175] For high-dimensional hyperspectral remote sensing images, there is a large amount of redundant information. If the input data dimension is too large, the complexity of the model will be increased. Therefore, reducing the dimension to the appropriate ...

Embodiment 2

[0179] Embodiment 2: In order to further verify the validity of the method proposed by the present invention, the PaviaUniversity and Indian Pines data sets will be used for verification below, and the classification method SAE-3DDRN described in the present invention will be combined with some traditional hyperspectral classification methods Compared with the current mainstream algorithms, the number of training samples used by each classification method is exactly the same.

[0180] Table 5 Classification accuracy of different classification methods on the Indian Pines dataset

[0181]

[0182]

[0183] right image 3 The hyperspectral image of Indian Pines shown is classified, and 10% samples are randomly selected from each type of ground object as the training set, and the remaining samples are used as the test set. On this data set, the classification method SAE-3DDRN proposed by the present invention uses a 5-layer stacked autoencoder structure to reduce the dimen...

Embodiment 3

[0187] Embodiment 3: In order to verify the impact of the classification method of the present invention on the hyperspectral classification accuracy under different sample ratios, this paper sets four training sample ratios of 5%, 10%, 15%, and 20%, respectively from Indian Pines and PaviaUniversity data sets are randomly selected, and the average and standard deviation of ten classification results are taken for each experimental result. Figure 10 , Figure 11 They are the point-line diagrams of OA, AA, and Kapp a changes under different training sample ratios of the method SAE-3DDRN described in this paper under the Indian Pines and Pavia University datasets. From Figure 10 , Figure 11 It can be seen that the SAE-3DDRN method proposed in this paper shows the best classification accuracy in OA, AA, and Kappa under different proportion samples of the two datasets. As the training sample size increases, the classification accuracy of most classification methods continues...

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 relates to a hyperspectral image classification method based on an auto-encoder and a 3D deep residual network. Compared with the prior art, the defect that remote sensing hyperspectralimage classification is difficult to carry out is overcome. The method comprises the following steps: obtaining a training sample; preprocessing the hyperspectral remote sensing image data to be trained; building and training a stack auto-encoder neural network model; constructing and training a 3D deep residual network; acquiring a hyperspectral remote sensing image to be classified; carrying outpreprocessing and dimensionality reduction of the hyperspectral remote sensing images to be classified; and obtaining a hyperspectral remote sensing image classification result. According to the method, a stack auto-encoder neural network model is built, dimensionality reduction is carried out on an original hyperspectral remote sensing image, and redundant information is eliminated; a residual error network module is introduced through a designed 3D convolutional neural network to properly increase the depth of the network, a 3D deep residual error network is established, space-spectrum joint information of a hyperspectral remote sensing image is extracted more effectively, and the problems of gradient disappearance and network degradation are avoided.

Description

technical field [0001] The invention relates to the technical field of hyperspectral image processing, in particular to a hyperspectral image classification method based on an autoencoder and a 3D deep residual network. Background technique [0002] With the continuous and rapid development of hyperspectral remote sensing technology at home and abroad, it has been widely used in agriculture, environmental science, and ground object observation. Hyperspectral remote sensing image data is a three-dimensional data cube containing rich spectral and spatial information. It has information on hundreds of continuous spectral segments of surface objects, which greatly improves the ability to identify and distinguish various types of ground objects. How to make full use of the high-resolution spatial spectrum information of hyperspectral data and continuously improve the classification accuracy has become the goal that researchers are constantly pursuing. However, hyperspectral imag...

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
IPC IPC(8): G06K9/00G06K9/62G06N3/04G06N3/08
CPCG06N3/084G06V20/194G06V20/13G06N3/045G06F18/213G06F18/214G06F18/24
Inventor 赵晋陵胡磊黄林生黄文江梁栋徐超张东彦翁士状郑玲
Owner ANHUI UNIVERSITY
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
Try Eureka
PatSnap group products

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

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap