[Two endmember extraction algorithms with combined spatial and spectral domain TM image]

Area of Science:

• Remote Sensing
• Image Analysis
• Spectral Unmixing

Context:

• Limited spectral bands and information in TM remote sensing images pose challenges for accurate endmember extraction.
• Existing methods may struggle with large regional scales and complex object compositions.

Purpose:

• To develop and compare two novel algorithms for endmember extraction from multispectral remote sensing data.
• To evaluate the trade-offs between speed, accuracy, and prior knowledge requirements for different extraction approaches.

Summary:

• The spatial split endmember extraction algorithm divides images into blocks based on object complexity and uses the hourglass algorithm.
• The region continuity algorithm also uses block division but incorporates homogenous object extraction/classification and spectral correlation matching.
• The spatial split algorithm is faster with less prior knowledge but risks extraction errors; the region continuity algorithm offers higher precision but requires more prior knowledge and is slower.

Impact:

• Both developed algorithms effectively address the challenge of endmember extraction in large-scale multispectral remote sensing.
• These methods demonstrate broad applicability and potential for advancing spectral unmixing techniques in various Earth observation applications.