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

A Hyperspectral Remote Sensing Image Classification Method Based on Dual Attention Mechanism

A hyperspectral remote sensing and attention technology, applied in the field of hyperspectral remote sensing images, can solve the problems of unsatisfactory, limited classification accuracy of training samples of hyperspectral remote sensing images, etc., to achieve good performance and reduce data redundancy.

Active Publication Date: 2022-02-01
ANHUI UNIVERSITY
View PDF14 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0009] The purpose of the present invention is to solve the defects of limited hyperspectral remote sensing image training samples and unsatisfactory classification accuracy in the prior art, and provide a hyperspectral remote sensing image classification method based on a double attention mechanism to solve the above problems

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
  • A Hyperspectral Remote Sensing Image Classification Method Based on Dual Attention Mechanism
  • A Hyperspectral Remote Sensing Image Classification Method Based on Dual Attention Mechanism
  • A Hyperspectral Remote Sensing Image Classification Method Based on Dual Attention Mechanism

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0180] Embodiment 1: In order to verify the impact of different parameter settings on the classification accuracy in the method proposed by the present invention (HDDA), the input sample window size w, the learning rate lr, and the Dropout parameters were mainly analyzed. In this embodiment, 5%, 1%, and 1% of the training samples are randomly selected from each type of ground objects in the IP, UP, and SA data sets to train the model, and the rest are used as test samples. In addition, the dimensionality reduction dimension d of the present invention is set to 10, and the HDDA network model is trained by using the MSE loss function and the Adam optimization algorithm. The batch size is uniformly 64, and the iteration is 200 times. Figure 5 (a) is the influence of the method HDDA of the present invention on the classification accuracy under different window sizes w. If the w value is small, the receptive field of the convolution kernel feature extraction will be insufficient, ...

Embodiment 2

[0184] Embodiment 2: In order to further verify the effectiveness of the algorithm of the present invention, this embodiment adopts three kinds of data sets of IP, UP, and SA for verification, and randomly selects 5%, 1%, and 1% from each type of ground object as training samples set, and the rest as the test sample set. And use REF-SVM, 3D-CNN, HybridSN, SSRN and R-HybridSN five hyperspectral image classification methods as the comparative experimental objects, the classification results take the mean of ten experimental results, and record the standard deviation, so as to verify the HDDA method classification performance.

[0185] Table 3. Classification accuracy of different classification methods on the IP dataset.

[0186]

[0187] right figure 2The shown IP hyperspectral image is classified, and the HDDA classification method proposed by the present invention uses a 5-layer stacked autoencoder structure to reduce the dimensionality of the hyperspectral remote sensi...

Embodiment 3

[0194] Embodiment 3: In order to further verify the classification performance of the HDDA classification method of the present invention under limited training samples, this embodiment sets different training sample ratios on the IP, UP, and SA data sets respectively, and sets respectively on the IP data set 2%, 4%, 6%, 8%, 10% five sets of training sample proportions, set 0.2%, 0.4%, 0.6%, 0.8% four sets of training sample proportions on the UP and SA data sets, by further reducing the training samples , so as to compare the overall classification accuracy of the classification method of the present invention and the comparative classification method.

[0195] Figure 10 Influence of HDDA on hyperspectral image classification under different sample ratios for the method described in the present invention. from Figure 10 It can be seen that the HDDA classification method proposed by the present invention shows the best classification accuracy in OA and OA under different p...

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 remote sensing image classification method based on a double-attention mechanism, which solves the defects of limited hyperspectral remote sensing image training samples and unsatisfactory classification accuracy compared with the prior art. The invention comprises the following steps: acquisition of training samples; construction and training of a stack autoencoder; construction of a mixed dense network; training of the mixed dense network; acquisition of samples to be classified and dimensionality reduction processing; The present invention can also obtain ideal hyperspectral remote sensing image classification results under the condition of less training samples.

Description

technical field [0001] The invention relates to the technical field of hyperspectral remote sensing images, in particular to a hyperspectral remote sensing image classification method based on a double attention mechanism. Background technique [0002] Hyperspectral remote sensing images are rich in information in both spectral and spatial dimensions, and have been widely used in environmental science, agriculture, and land cover mapping. Hyperspectral images contain dozens or even hundreds of continuous spectral bands, which are time-consuming and laborious to label, and the number of available training samples is usually limited. Therefore, due to the hyperspectral "small sample problem", it is still a challenging task to achieve high-precision classification of hyperspectral images. [0003] There is a large amount of redundant information between adjacent bands of hyperspectral images, which will cause a decrease in storage and computing efficiency during subsequent cla...

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 Patents(China)
IPC IPC(8): G06V30/19G06V20/13G06N3/04G06N3/08
CPCG06N3/08G06V20/13G06N3/045G06F18/2411G06F18/214
Inventor 赵晋陵胡磊黄林生梁栋徐超黄文江翁士状张东彦郑玲
Owner ANHUI UNIVERSITY
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

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