研究目的
To create a publicly available dataset with ground-truth for evaluating and comparing the performance of hyperspectral unmixing algorithms.
研究成果
The study successfully creates a preliminary publicly available dataset for evaluating spectral unmixing algorithms, validated with both linear and nonlinear methods. Future work will aim to enrich the dataset with more scenes to mimic different practical cases.
研究不足
The lab-made dataset may not fully replicate the properties of real airborne data, limiting its direct applicability to real-world scenarios.
1：Experimental Design and Method Selection:
The study involves creating controlled experimental scenes in a laboratory setting to generate hyperspectral datasets with known pure material spectra and compositions.
2：Sample Selection and Data Sources:
The dataset includes checkerboard-type data and mixed granules, with known mixture fractions and patterns.
3：List of Experimental Equipment and Materials:
The hyperspectral data is collected using the GaiaSorter and GaiaField system, with specific parameters for moving speed, spectral resolution, spatial resolution, distance between lens and samples, and exposure time.
4：Experimental Procedures and Operational Workflow:
Data preprocessing includes black-white normalization to remove dark current effects and uneven light intensity.
5：Data Analysis Methods:
The study applies linear (FCLS, NCLS) and nonlinear (K-Hype) unmixing algorithms to the dataset to evaluate their performance.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
