<title>Statistical parametric signature/sensor/detection model for multispectral mine target detection</title>

Area-based results for mine detection

“…There have been several studies of mine detection using statistical methods and involving different sensors, including [1]- [7], [12], [14]- [19]. In the case of the direct point-by-point exhaustive detection and search for mines, which is directly related to the present work, current detection techniques do not provide sufficiently high performance results, especially in cases where the minefield is heavily cluttered [9], [10].…”

“…Whenever a mine or a false alarm is located at some point , the set of points centered at will be considered to have been explored and will be removed from further consideration concerning false-alarm rates. By "point" , we mean some location in the minefield, assuming that the whole field has been discretized in 1 1 unit squares, so that point (0, 0) would in fact refer to the square contained in the vertices (0, 0), (0, 1), (1, 0), (1,1). Note that if some point has been included previously in the declaration area of some other point where a mine or false alarm has previously been detected, then should not be eliminated twice from the areas being scanned.…”

Spectral imaging system analytical model for subpixel object detection

“…Since detector parameters are often specified in terms of electrons, the noise terms are summed in a root sum squared sense in that domain before being converted to noise equivalent spectral radiance. To calculate the photon noise term , the input radiance is converted to electrons and the square root taken as shown in (11). Note that this calculation is done for each object and background signal and for each spectral channel separately, with appropriate spectral variation for (11) where input spectral radiance [mW/(cm -sr-m)]; system throughput (cm -sr); optical transmittance; quantum efficiency; integration time (in seconds); spectral channel bandwidth ( m); spectral channel central wavelength ( m); Planck's constant J/s; the speed of light m/s; unit conversion constant .…”

“…To calculate the photon noise term , the input radiance is converted to electrons and the square root taken as shown in (11). Note that this calculation is done for each object and background signal and for each spectral channel separately, with appropriate spectral variation for (11) where input spectral radiance [mW/(cm -sr-m)]; system throughput (cm -sr); optical transmittance; quantum efficiency; integration time (in seconds); spectral channel bandwidth ( m); spectral channel central wavelength ( m); Planck's constant J/s; the speed of light m/s; unit conversion constant . The total detector noise (in electrons) is then calculated as the root-sum-square of the photon noise , the thermal noise , and the multiplexer/readout noise as follows:…”

“…This modeling approach uses statistical representations of various land cover classes and analytically propagates them through the remote sensing system. A similar approach was adopted by Schwartz et al [11], to analyze multispectral sensor systems in mine detection applications. These statistical approaches contrast with image simulation models such as HySIM [12], [13] and DIRSIG [14], which produce synthetic images suitable for analysis with 0196 traditional tools.…”