Related Experiment Video
Updated: Aug 16, 2025
05:56
Objectification of Tongue Diagnosis in Traditional Medicine, Data Analysis, and Study Application
Published on: April 14, 2023
2.6K
Semantic Segmentation of Hyperspectral Remote Sensing Images Based on PSE-UNet Model
Jiaju Li1, Hefeng Wang1,2, Anbing Zhang1,2
1School of Mining and Geomatics Engineering, Hebei University of Engineering, Handan 056038, China.
Sensors (Basel, Switzerland)
|December 23, 2022
Summary
This study introduces PSE-UNet for hyperspectral remote sensing image semantic segmentation, improving accuracy by combining principal component analysis (PCA) and attention mechanisms. The novel dataset splitting method ensures fair model comparison and addresses challenges in high-dimensional data.
Area of Science:
- Remote Sensing
- Deep Learning
- Computer Vision
Background:
- Deep learning, particularly convolutional neural networks (CNNs), is crucial for improving land cover classification accuracy in hyperspectral remote sensing images (HSRSI).
- Traditional dataset partitioning methods in HSRSI semantic segmentation can lead to information leakage, hindering fair model comparisons.
- Existing models face limitations due to small sample sizes and high dimensionality of HSRSI, with a lack of clear guidance on principal component analysis (PCA) for dimensionality reduction.
Purpose of the Study:
- To develop a robust semantic segmentation model for HSRSI that overcomes data challenges.
- To introduce a novel, fair dataset splitting strategy for HSRSI analysis.
- To investigate the impact of PCA dimensionality reduction and attention mechanisms on HSRSI semantic segmentation performance.
Main Methods:
- A non-overlapping sliding window strategy with a judgment mechanism was employed for hyperspectral dataset splitting, ensuring data integrity.
- A novel PSE-UNet model was designed, integrating PCA, an attention mechanism, and the UNet architecture for HSRSI semantic segmentation.
- The cumulative variance contribution rate (CVCR) was introduced as a metric for PCA dimensionality reduction to analyze the Hughes phenomenon.
Main Results:
- The proposed PSE-UNet model demonstrated superior performance compared to existing semantic segmentation algorithms on the Salinas dataset.
- The study analyzed factors influencing PSE-UNet's performance, providing insights into model optimization.
- The CVCR metric effectively addressed dimensionality reduction challenges and the Hughes phenomenon in HSRSI.
Conclusions:
- The PSE-UNet model offers a significant advancement in HSRSI semantic segmentation, outperforming current methods.
- The developed dataset splitting strategy provides a reliable benchmark for evaluating HSRSI models.
- This research offers valuable insights and a reference for future studies in hyperspectral image analysis and deep learning applications.
