研究目的
Investigating the effects of anomalies on linear spectral unmixing of hyperspectral images and developing a Bayesian algorithm for joint linear unmixing and anomaly detection.
研究成果
The proposed Bayesian algorithm for robust linear spectral unmixing demonstrates superior performance in terms of endmember and abundance estimation and anomaly detection compared to traditional methods. It effectively identifies structured anomalies, enhancing the analysis of hyperspectral images.
研究不足
The computational cost is higher than traditional methods like VCA-FCLS. The method's performance degrades as the proportion of outliers increases or when the variances of the noise and anomalies are similar.
1：Experimental Design and Method Selection:
A Bayesian algorithm is proposed for linear spectral unmixing that accounts for anomalies. The model assumes pixel reflectances are linear mixtures of unknown endmembers, corrupted by an additional nonlinear term modeling anomalies, and additive Gaussian noise. A Markov random field is used for anomaly detection based on spatial and spectral structures.
2：Sample Selection and Data Sources:
Synthetic and real hyperspectral images are used to evaluate the algorithm's performance.
3：List of Experimental Equipment and Materials:
Hyperspectral images acquired by AVIRIS satellite and Hyspex hyperspectral scanner.
4：Experimental Procedures and Operational Workflow:
The algorithm involves estimating the endmember matrix, abundance matrix, noise variances, and outlier matrix from the observation matrix using a hierarchical Bayesian model and a sampling method.
5：Data Analysis Methods:
The performance is evaluated using root normalized mean square error (RNMSE) for abundance estimation and spectral angle mapper (SAM) for endmember estimation.
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
