Related Experiment Video
Updated: Nov 10, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.1K
Multi-U-Net: Residual Module under Multisensory Field and Attention Mechanism Based Optimized U-Net for VHR Image
Si Ran1,2, Jianli Ding1,2, Bohua Liu1,2
1Key Laboratory of Smart City and Environment Modeling of Autonomous Region Universities, College of Resources and Environment Sciences, Xinjiang University, Urumqi 830046, China.
Sensors (Basel, Switzerland)
|April 3, 2021
Summary
This study introduces a novel residual module under a multisensory field (RMMF) to enhance U-Net for very high resolution (VHR) image semantic segmentation. The RMMF module significantly improves feature extraction from multiscale data in airborne and spaceborne imagery.
Area of Science:
- Computer Vision
- Machine Learning
- Remote Sensing
Background:
- Very high resolution (VHR) image acquisition is increasing, presenting challenges for traditional semantic segmentation.
- Convolutional Neural Networks (CNNs), like U-Net, are effective for image interpretation but can be improved for multiscale feature extraction.
Purpose of the Study:
- To develop an improved network module for enhanced multiscale feature extraction in VHR images.
- To address limitations in U-Net's full-scale information utilization.
Main Methods:
- Constructed a residual module under a multisensory field (RMMF) to extract multiscale features.
- Integrated an attention mechanism to optimize feature information.
- Replaced standard convolutional layers in U-Net with the RMMF module.
Main Results:
- The RMMF module effectively extracts multiscale features using parallel convolutional layers and residual blocks.
- Experiments on Gaofen-2 and Potsdam datasets demonstrated superior performance compared to existing methods.
- The enhanced U-Net model showed improved results in airborne and spaceborne image segmentation.
Conclusions:
- The proposed RMMF module is a universal and extensible solution for improving CNN-based semantic segmentation.
- This approach offers better performance for VHR image interpretation, particularly in remote sensing applications.
