Progressive Semantic Enhancement Network for Hyperspectral and LiDAR Classification.

Area of Science:

• Remote Sensing
• Computer Vision
• Geospatial Analysis

Background:

• Joint classification of hyperspectral image (HSI) and light detection and ranging (LiDAR) data offers enhanced accuracy.
• Integrating spectral information from HSI and elevation data from LiDAR presents a significant challenge in multimodal fusion.

Purpose of the Study:

• To propose a novel progressive semantic enhancement network (PSENet) for accurate HSI and LiDAR data classification.
• To effectively fuse spatial, spectral, and elevation information using a progressive joint spatial-spectral attention mechanism.

Main Methods:

• Developed PSENet with a spatial grouping constraint (SAGC) module for multiscale spatial feature extraction.
• Incorporated a spectral weighting constraint (SEWC) module to enhance semantic features in the spectral dimension.
• Employed a progressive approach to gradually enhance feature extraction through spatial and spectral constraint modules.

Main Results:

• PSENet demonstrated superior performance compared to state-of-the-art methods across three benchmark datasets.
• The SAGC and SEWC modules effectively integrated spatial, spectral, and elevation information.
• Achieved more accurate classification of ground objects by leveraging multimodal data fusion.

Conclusions:

• PSENet provides a promising approach for accurate classification by effectively fusing HSI and LiDAR data.
• The proposed spatial and spectral constraint modules are key to successful multimodal data integration.
• The developed method advances the field of remote sensing data classification.