|
|
Technology Matched Filter Detector (MFD) Detecting imagery. Many analysis paradigms have been derived to address this need. One commonly used detector is the Matched Filter Detector (MFD). The MFD is derived under the assumption that the non-target components (i.e., structured interference, S) in a mixed pixel are orthogonal to the desired target signature, H. This idea, using a vector subspace analog, is illustrated in figure 1. Unfortunately there is a problem with the MFD when it is applied to spectra that arise from mixed pixels; and that is that the interference subspace is usually not orthogonal to the target subspace. In this case, the interference can "leak" into the output of the MFD as shown in figure 2. This additional energy can increase the matched filter score and cause false indication of the presence of the target. The problem of leakage can be addressed by incorporating interference into a modified form of the MFD called the Matched Subspace Detector (MSD). The MSD projects the measurement vector onto a new space, G, which is the projection of the target signature into the space orthogonal to the background spectra. These projections are illustrated in figure 3. The projection onto the space perpendicular to the background effectively nulls any contributions from the background which gives the MSD its invariance to interference shown in figure 4. In this case, 210 band HYDICE data were used to detect a series of manmade objects with known pixel locations. Data Fusion Corporation |
