Related Experiment Video
Updated: Jul 30, 2025
05:56
Objectification of Tongue Diagnosis in Traditional Medicine, Data Analysis, and Study Application
Published on: April 14, 2023
2.6K
A deep learning method for optimizing semantic segmentation accuracy of remote sensing images based on improved UNet.
Xiaolei Wang1,2,3, Zirong Hu4, Shouhai Shi4
1The School of Geoscience and Technology, Zhengzhou University, Zhengzhou, 450001, China. xiaolei8788@zzu.edu.cn.
Scientific Reports
|May 10, 2023
Summary
Adaptive Feature Fusion UNet (AFF-UNet) improves remote sensing imagery semantic segmentation. This model enhances accuracy and object integrity by adaptively fusing features and incorporating attention mechanisms.
Area of Science:
- Computer Science
- Remote Sensing
- Artificial Intelligence
Background:
- Semantic segmentation of remote sensing imagery (RSI) is crucial but challenging due to diverse landscapes and object sizes.
- Existing models struggle with accurately segmenting varied geo-objects in RSI.
Purpose of the Study:
- To propose an optimized convolutional network, Adaptive Feature Fusion UNet (AFF-UNet), for improved semantic segmentation of RSI.
- To enhance the accuracy and integrity of segmented objects in remote sensing data.
Main Methods:
- Developed AFF-UNet featuring dense skip connections and an adaptive feature fusion module.
- Incorporated channel and spatial attention modules to capture inter-channel and inter-position relationships.
- Evaluated AFF-UNet on two public RSI datasets, comparing it with existing models.
Main Results:
- AFF-UNet achieved a 1.09% increase in average F1 score and 0.99% improvement in overall accuracy on the Potsdam dataset compared to DeepLabv3+.
- Qualitative results showed reduced class confusion and improved segmentation of objects with varying sizes.
- Demonstrated enhanced object integrity in the segmented RSI.
Conclusions:
- The proposed AFF-UNet model significantly optimizes the accuracy of semantic segmentation for remote sensing imagery.
- AFF-UNet's adaptive feature fusion and attention mechanisms contribute to superior performance in complex RSI segmentation tasks.
