Related Experiment Video
Updated: Jun 13, 2026
03:31
End-To-End Deep Neural Network for Salient Object Detection in Complex Environments
Published on: December 15, 2023
Refined Deformable-DETR for Electric Pylon Detection Based on Optical Satellite Image
Jun Yang1,2, Yu Sun1,2, Yingjun Zhao1,2
1Beijing Research Institute of Uranium Geology, Beijing 100029, China.
Sensors (Basel, Switzerland)
|June 12, 2026
Summary
This study introduces a new method for automatically detecting electric pylons in remote sensing images, improving accuracy in complex environments. The refined framework enhances object query representations for better powerline monitoring.
Area of Science:
- Computer Vision
- Remote Sensing Technology
- Artificial Intelligence
Background:
- Automatic electric pylon detection is crucial for powerline monitoring.
- Challenges include complex backgrounds, small targets, and variable pylon-shadow structures.
Purpose of the Study:
- To enhance electric pylon detection in optical remote sensing imagery.
- To improve the accuracy and robustness of Transformer-based detection models.
Main Methods:
- Proposed a Refined Deformable-DETR framework.
- Introduced a Spatial Context-aware Query Modulation (SCQM) module.
- SCQM aggregates image context and recalibrates object queries.
Main Results:
- Improved Average Precision (AP) from 72.7% to 74.1% on the EPRD dataset.
- Enhanced APs (small objects) from 47.2% to 50.9% on the EPRD dataset.
- Demonstrated generalization capability on the public EPD dataset.
Conclusions:
- Context-aware query modulation effectively enhances Transformer-based electric pylon detection.
- The SCQM module improves performance in complex remote sensing scenarios.
- The proposed method offers a significant advancement for powerline infrastructure monitoring.