A new strategy based on multi-source remote sensing data for improving the accuracy of land use/cover change...

Area of Science:

• Remote Sensing
• Geospatial Analysis
• Environmental Monitoring

Background:

• Land Use/Cover Change (LUCC) is vital for sustainable land management but current feature extraction methods struggle with complex data, hindering accurate land cover categorization.
• Efficiently extracting and classifying land cover data is essential for effective regional planning and environmental monitoring.

Purpose of the Study:

• To introduce novel feature extraction and dimensionality reduction algorithms for improved land cover classification.
• To evaluate the effectiveness of these methods using Unmanned Aerial Vehicle (UAV) LiDAR and Hyperspectral Imaging (HSI) data across diverse environments.
• To identify optimal feature subsets and assess the contribution of individual data sources and features to classification accuracy.

Main Methods:

• Developed a new feature extraction algorithm and a Segmented and Stratified Principal Component Analysis (SS-PCA) for dimensionality reduction.
• Applied methods to UAV LiDAR and HSI data from urban, agricultural, and natural environments, extracting 157 features.
• Utilized a random forest classifier with an optimal feature subset for land cover classification.

Main Results:

• The SS-PCA method enhanced feature bands, reduced HSI noise, and improved overall classification accuracy to 91.17%.
• The CFW method improved accuracy for LiDAR (78.10%), HSI (89.87%), and combined LiDAR+HSI data (97.17%).
• LiDAR intensity and HSI features were identified as critical for enhancing classification accuracy.

Conclusions:

• The proposed methods, particularly the integration of LiDAR and HSI data, significantly improve ground-level fine classification accuracy.
• These techniques effectively mitigate classification noise and offer a robust approach for diverse land cover categorization scenarios.
• The findings underscore the potential of advanced feature extraction and data fusion for precise land management and planning.