A remote sensing image scene classification method

By using improved SE module and bilinear pooling technology in remote sensing image classification, attention enhancement and feature fusion of features extracted by convolutional neural networks has been solved, and the problem that feature extraction in the existing technology is difficult to meet the high-precision needs, achieving a more efficient remote sensing image scene classification effect.

CN114266955BActive Publication Date: 2025-03-28Chinese People's Liberation Army Cyberspace Force Information Engineering University
0 Cites 0 Cited by

Patent Information

Application Number
CN202010975267.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2020-09-16
Publication Date
2025-03-28
Estimated Expiration
2040-09-16

AI Technical Summary

Technical Problem

The existing methods cannot extract features with strong discrimination, resulting in poor classification of remote sensing image scenes, especially in scenarios where there are large intra-class differences and high similarity between classes, it is difficult to achieve high-precision classification.

Method used

By inputting the remote sensing images to be classified into the trained convolutional neural network, the feature maps output from at least two convolutional layers are extracted and the feature maps are enhanced with the improved SE module. Then, the enhanced feature map is bilinear pooled and spliced ​​to generate the final feature vector for scene classification.

Benefits of technology

It improves the diversity and effectiveness of features, enhances the discrimination of features, and can more effectively distinguish fine-grained remote sensing image scenarios, significantly improving the recognition accuracy of similar categories.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The present invention provides a remote sensing image scene classification method, belonging to the field of image processing. The method comprises: inputting a remote sensing image to be classified into a trained convolutional neural network, and extracting feature maps output by at least two convolutional layers in the convolutional neural network, wherein one convolutional layer is the last convolutional layer of the convolutional neural network, and the size of each feature map is the same; using an improved SE module to perform attention enhancement on each feature map respectively; the improved SE module is obtained by adding a batch normalization layer after each fully connected layer of the SE module, and replacing all activation functions in the SE module with Hardtanh functions; performing bilinear pooling on each enhanced feature map in pairs to obtain a corresponding initial feature vector, and splicing each initial feature vector to obtain a final feature vector; and using the final feature vector to realize scene classification of the remote sensing image to be classified. The final feature of the present invention has strong distinguishability and can achieve higher recognition accuracy on scenes of similar categories.
Need to check novelty before this filing date? Find Prior Art