A simple semi-automatic approach for land cover classification from multispectral remote sensing imagery.

Area of Science:

• Earth and Space Sciences
• Remote Sensing
• Geographic Information Systems (GIS)

Background:

• Land cover data are crucial for scientific research, but satellite-based classification is challenging.
• Efficient and automated land cover classification methods are needed to overcome these challenges.

Purpose of the Study:

• To propose an automatic scheme for land cover classification using multispectral remote sensing images.
• To minimize human involvement in the classification process through change detection and semi-supervised learning.

Main Methods:

• Developed an automatic classification scheme using multispectral remote sensing images.
• Employed change detection and a semi-supervised classifier, requiring only prior land cover maps and existing images.
• Tested the method on Environment Satellite 1 (HJ-1) images in Shanghai, China.

Main Results:

• Achieved accurate land cover classification into five main types with a Kappa value of 0.79.
• Demonstrated statistical area biases below 6% in validation.
• Successfully classified land cover using prior maps and spectral features.

Conclusions:

• The proposed semi-automatic approach offers a simple yet accurate method for land cover classification.
• This technique is valuable for mapping areas with limited ground reference data or rapid land cover changes.
• The method integrates visual interpretation accuracy with automatic classification efficiency.