E. Lo (USA)
Anomaly detection, hyperspectral imaging, target detection, multivariate statistics.
An anomaly detector for detecting anomalies in a large natural background in real-time hyperspectral imaging is developed in this paper from a likelihood ratio test. The null hypothesis tests if the residual of a test pixel having partialled out the high-variance principal components using a linear predictor is from a multivariate normal distribution. The alternative hypothesis tests if the residual of the test pixel having partialled out the low-variance principal components using a linear predictor is from another multivariate normal distribution. The number of predictors for each spectral component of the test pixel can vary from one spectral component to another. The experimental results from a hyperspectral data cube show that the anomaly detector based on variable number of predictors outperforms the RX and SSRX detectors and is robust with respect to its parameter.
Important Links:
Go Back
