Effective Anomaly Space for Hyperspectral Anomaly Detection

Background-Annihilated Target-Constrained Interference-Minimized Filter (TCIMF) for Hyperspectral Target Detection

“…These experiments were very crucial because they showed that when the sufficient ground truth is necessary for HCEM, ECEM, KTCIMF and DS-BA-TCIMF and LRaSMD-TCIMF and CDASC-TCIMF to perform well as documented in Tables III and IV Specifically, the great advantage of BA-TICMF over the DL-based methods is that BA-TCIMF makes use of data sphering and unsupervised MXSVD to find unwanted BKG signatures through GoDec model. This has been shown by the experiments conducted for the San Diego airport scene in Section VIII in [38] where hCEM and ECEM were shown in [20] to outperform machine learning-based methods in BKG suppression such as PTA (prior-based tensor approximation) in [39], LSDM-MoG (low-rank and sparse decomposition with mixture of Gaussians) in [40] and LRR (low rank representation) in [41] VII. CONCLUSION Target-constrained interference-minimized filter (TCIMF) is an effective subpixel target detector in many applications.…”

“…onto Ω via the projection P Ω . For various comparative analyses can be found in references [14][15]17,30,38], specifically, reference [38]. Those who are interested in Scare encouraged to consult [17,38] for more details of SC interpretations and derivations In particular, Chang in [38] provided how the concept of SC was derived and also offered different versions of SC.…”

“…The location of these planes can be obtained from the ground truth in Fig. 4(c) and then be used as the desired signature d. To estimate VD, HFC and its noise-whitened HFC (NWHFC) [1,19,38] were used to estimate p. This was followed by MXSVD to estimate j. Table I tabulates the value of p and j estimated by PF ≤ 10 -3 and MX-SVD so that m = p -j and k = j × N and parameter setting for KTCIMF.…”

“…As you may be noticed that TD in (37), BS in (38), TDBS in (39), and SNPR in (40) all only considered the two 2D ROC curves. This section derives a new detection measure by taking into account AUC values of its corresponding three 2D ROC curves, 2D ROC curves of (PD, PF), 2D ROC curves of (PD,τ), 2D ROC curves of (PF,τ), named overall detection probability (ODP), denoted by AUCODP, as…”

Target-to-Anomaly Conversion for Hyperspectral Anomaly Detection

“…In recent years, this approach has been a major trend in anomaly detection. Of particular interest is the LRaSMD model that has been shown in [29], [30], [31], [32], [33], and [34] to take care of BS and noise effectively.…”