Progressive Band Processing of Anomaly Detection in Hyperspectral Imagery

Abstract

Anomalies are generally unknown and unexpected and cannot be detected with prior knowledge. Consequently, it is highly desirable to have them detected in an unsupervised manner on a timely basis. One way to do so is to perform anomaly detection while the process of data collection is still ongoing, so that weak anomalies will not be dominated and overwhelmed by subsequent detected strong anomalies. This paper presents an approach to progressive band processing of anomaly detection (PBP-AD) band by band according to band sequential (BSQ) format. In other words, anomaly detection can be carried out band by band progressively without waiting for entire bands completely acquired. This significant advantage allows anomaly detection to be implemented in real time in the sense of progressive band processing with the data processing taking place and data being collected at the same time. This capability also paves a way for anomaly detection in future satellite data communication and transmission where the data can be processed and down linked from satellites band by band simultaneously. Index Terms-Anomaly detection, band sequential (BSQ), causal band K-RXD (CBK-RXD), causal band R-RXD (CBR-RXD), causal band sample correlation matrix (CBSCM), K-RXD, PBP-K-RXD, PBP-R-RXD, R-RXD, progressive band processing (PBP), progressive band processing of anomaly detection (PBP-AD).

Progressive Sample Processing of Band Selection for Hyperspectral Image Transmission

“…First, we can observe all the BS results in the whole transmission, where the local BS results (i.e., the BS results in a particular time segment) would not be missed. This is one of the most attractive features in progressive processing [34,35]. For instance, some "weaker bands" would appear during the transmission and disappear later on.…”

“…This issue has not been emphasized in the related literature. Unlike the PBP works [33][34][35], the PSP-OMPBS algorithm is further involved with the iterations related to sequential search. The computing time may increase with the drastic increase of p. Thus, the computation time must be further reduced to satisfy the prerequisite.…”

“…Unlike related work about PBP [33][34][35], where the recursive calculation can be applied to each iteration of PBP, Equation (9) of Re-OMPBS can only be implemented under particular conditions. To understand it, we assume that Ω j (n) presents the BS result at the jth iteration in the nth stage, and the result was just obtained.…”

“…PBP assumes that the HSI data is transmitted by band sequential (BSQ) format, and processes the image immediately after receiving a new band. Under the PBP framework, many common topics, such as spectral unmixing, target detection, and anomaly detection could be realized [33][34][35]. Unlike the traditional algorithms, which can only be implemented on the collected data, PBP makes use of recursive processing, to instantly process the current data segment.…”

Recursive Band Processing of Automatic Target Generation Process for Finding Unsupervised Targets in Hyperspectral Imagery

“…Another important note is that RBP-ATGP is also very different from progressive band processing of anomaly detection (PBP-AD) in [26] and progressive band processing of constrained energy minimization (PBP-CEM) in [27] in the sense that the recursive equations used by RBP-ATGP are derived from the target signature matrix U in (1), while the recursive equations used by PBP-AD and PBP-CEM are derived from the autocorrelation matrix R formed by data sample vectors, not target signal sources. In particular, the recursive equation (8) derived for RBP-ATGP iterates the undesired signatures in U via (6) which has nothing to do with data sample vectors.…”