TY - GEN
T1 - Hyperspectral Target Detection Using Segmented Matched Filter with Local Covariance Reassignment
AU - Elisha, Haim
AU - Rotman, Stanley
N1 - Publisher Copyright:
© 2022 IEEE.
PY - 2022/1/1
Y1 - 2022/1/1
N2 - Segmentation is useful during sub-pixel target detection in hyperspectral data. Our standard algorithm uses estimates of the actual pixel being examined (based on surrounding pixels) and of the covariance matrix of the background (traditionally based on all the pixels in the image). Previous works have showed that improving performance in sub-pixel target detection can be achieved by making better estimates of the covariance matrix by using segmentation. One of the challenges is that pixel assignment to its segment can be influenced by the presence of the target that will lead us to miss the target. Therefore, it is desirable to assign the examined pixel by using the neighbors of the pixel assignment without involving the pixel itself is needed. We developed a new reassignment segmentation without involving the central pixel. Using simulations and several analytical tools, we analyzed the matched-filter algorithm, both with and without segmentation, and compare performances of the receiver operating characteristic curves. Our algorithm showed better receiver operating characteristic curves in low false positive rate in the range 0-0.01 (the operating range of our applications), i.e., we got a higher true positive rate for the same false positive rate.
AB - Segmentation is useful during sub-pixel target detection in hyperspectral data. Our standard algorithm uses estimates of the actual pixel being examined (based on surrounding pixels) and of the covariance matrix of the background (traditionally based on all the pixels in the image). Previous works have showed that improving performance in sub-pixel target detection can be achieved by making better estimates of the covariance matrix by using segmentation. One of the challenges is that pixel assignment to its segment can be influenced by the presence of the target that will lead us to miss the target. Therefore, it is desirable to assign the examined pixel by using the neighbors of the pixel assignment without involving the pixel itself is needed. We developed a new reassignment segmentation without involving the central pixel. Using simulations and several analytical tools, we analyzed the matched-filter algorithm, both with and without segmentation, and compare performances of the receiver operating characteristic curves. Our algorithm showed better receiver operating characteristic curves in low false positive rate in the range 0-0.01 (the operating range of our applications), i.e., we got a higher true positive rate for the same false positive rate.
KW - Covariance matrix
KW - False positive rate (FPR)
KW - Hyperspectral
KW - Matched filter
KW - Receiver operation characteristic (ROC) curve
KW - Segmentation
KW - Subpixel target detection
KW - True positive rate (TPR)
UR - http://www.scopus.com/inward/record.url?scp=85143130878&partnerID=8YFLogxK
U2 - 10.1109/WHISPERS56178.2022.9955083
DO - 10.1109/WHISPERS56178.2022.9955083
M3 - Conference contribution
AN - SCOPUS:85143130878
T3 - Workshop on Hyperspectral Image and Signal Processing, Evolution in Remote Sensing
BT - 2022 12th Workshop on Hyperspectral Imaging and Signal Processing
PB - Institute of Electrical and Electronics Engineers
T2 - 12th Workshop on Hyperspectral Imaging and Signal Processing: Evolution in Remote Sensing, WHISPERS 2022
Y2 - 13 September 2022 through 16 September 2022
ER -