Related Experiment Video
Updated: Dec 11, 2025
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
876
Low Dimensional Discriminative Representation of Fully Connected Layer Features Using Extended LargeVis Method for
1State Key Laboratory of Virtual Reality Technology and Systems, School of Computer Science and Engineering, Beihang University, Beijing 100191, China.
Sensors (Basel, Switzerland)
|August 23, 2020
Summary
A new E-LargeVis method enhances high-resolution remote sensing image retrieval by reducing feature dimensionality. This approach improves accuracy and speed, outperforming existing dimensionality reduction techniques.
Area of Science:
- Computer Science
- Remote Sensing
- Artificial Intelligence
Background:
- High-resolution remote sensing image retrieval is crucial for data management.
- Effective feature extraction requires low-dimensional, representative, and discriminative features.
- Dimensionality reduction is key to improving feature quality and retrieval performance.
Purpose of the Study:
- To propose an extended LargeVis (E-LargeVis) method for dimensionality reduction in high-resolution remote sensing image retrieval.
- To develop a novel retrieval method that generates stronger representative and discriminative deep features.
- To enhance the accuracy and speed of remote sensing image retrieval systems.
Main Methods:
- Utilized a channel attention-based ResNet50 as a backbone network to extract fully connected layer features.
- Applied the proposed E-LargeVis method, incorporating support vector regression, for dimensionality reduction.
- Employed L2 distance for similarity measurement to facilitate image retrieval.
Main Results:
- The E-LargeVis method effectively reduced the dimensionality of high-dimensional features.
- The proposed method achieved improved retrieval performance across four benchmark datasets (UCM, RS19, RSSCN7, AID).
- E-LargeVis significantly outperformed other dimensionality reduction techniques in retrieval accuracy and efficiency.
Conclusions:
- The E-LargeVis method offers a significant advancement in high-resolution remote sensing image retrieval.
- This approach provides a robust solution for obtaining discriminative low-dimensional features.
- The findings demonstrate the potential of E-LargeVis for various convolutional neural network architectures in remote sensing applications.
