Related Experiment Video
Updated: Sep 18, 2025
06:54
Photorealistic Learned Landscapes for Augmented Reality
Published on: June 27, 2025
196
Building extraction from remote sensing images based on multi-scale attention gate and enhanced positional
Rui Xu1, Renzhong Mao1, Zhenxing Zhuang1
1School of Computer Science and Mathematics, Fujian University of Technology, Fuzhou, Fujian, China.
Peerj. Computer Science
|June 26, 2025
Summary
This study introduces a new deep learning method for precise building extraction from remote sensing images. The novel approach enhances edge accuracy and detail, outperforming existing models on benchmark datasets.
Area of Science:
- Remote Sensing
- Computer Vision
- Geospatial Analysis
Background:
- Building extraction from high-resolution remote sensing images is a key research area.
- Current deep learning methods face challenges with blurred edges, incomplete structures, and detail loss.
Purpose of the Study:
- To develop a novel building extraction method for accurate contours and clear boundaries.
- To improve the precision of building shape and edge information extraction.
Main Methods:
- Utilized U-Net as the primary framework.
- Introduced a multi-scale attention gate module in the encoder for improved multi-scale information capture.
- Implemented a decoder module to enhance positional information for precise localization.
Main Results:
- The proposed method demonstrated consistent performance improvements across three benchmark datasets (Massachusetts, WHU, Inria).
- Achieved significant increments in intersection over union (IoU) metrics compared to six state-of-the-art models.
- Showcased effective integration of multi-scale features and optimized building edges.
Conclusions:
- The novel method significantly enhances building extraction accuracy and edge definition.
- The approach offers superior performance over existing methodologies in remote sensing building extraction.
- Validated effectiveness through comprehensive evaluations on multiple benchmark datasets.
