A hyperspectral image classification method fusing stacked auto-encoding network and CNN

By integrating stacked autoencoder networks and CNNs, hyperspectral remote sensing images are preprocessed and pre-trained. By fine-tuning and optimizing network parameters, efficient classification of hyperspectral remote sensing images is achieved, solving the problem of separation between dimensionality reduction and classification, improving classification accuracy and simplifying the process.

CN115471737BActive Publication Date: 2026-07-21NORTHWEST A & F UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
NORTHWEST A & F UNIV
Filing Date
2021-05-25
Publication Date
2026-07-21

AI Technical Summary

Technical Problem

In existing hyperspectral remote sensing image classification methods, the dimensionality reduction and classification processes are separated, resulting in features that are not suitable for classification. Furthermore, traditional methods have low classification accuracy and complex workflows.

Method used

A hyperspectral remote sensing image classification algorithm that integrates a stacked autoencoder network and a CNN is adopted. By pre-filtering and standardizing the hyperspectral remote sensing images, pre-training the stacked autoencoder network, and then connecting it with the CNN network and fine-tuning it, the network parameters are optimized to achieve data dimensionality reduction and classification in one go.

Benefits of technology

It improves the classification accuracy of hyperspectral remote sensing images, simplifies the workflow, reduces computational complexity, and still achieves good results with a small number of training samples, solving the problem of difficulty in obtaining real ground cover labels.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure QLYQS_1
    Figure QLYQS_1
  • Figure 210525113225
    Figure 210525113225
  • Figure 210525113234
    Figure 210525113234
Patent Text Reader

Abstract

The application discloses a hyperspectral remote sensing image classification method based on a fusion stack auto-encoding network and a CNN, and steps are as follows: (1) performing Gaussian pre-filtering on the hyperspectral remote sensing image; (2) performing layer-by-layer greedy pre-training on the stack auto-encoding network to complete initialization of the encoding network; and (3) connecting the encoding network with the CNN network, performing classification training, realizing Fine-Tuing of the overall network, simultaneously optimizing the encoder network and the CNN network, optimizing the overall effect of the network, and improving classification precision. The application combines the advantages of the auto-encoding network and the CNN network, changes a traditional two-step independent process of 'dimension reduction first and then classification', combines the data dimension reduction process with the data classification process, simplifies the work flow of the hyperspectral remote sensing image classification, and achieves better precision in the process of the hyperspectral remote sensing image classification, thereby providing a new idea for the hyperspectral remote sensing image classification.
Need to check novelty before this filing date? Find Prior Art