Knowing the sensed area of the target surface covered by the instrument FOV with a certain
shooting distance is important for the experiments of spectral mixture. Ideally, according to the
calibrated FOV diagram provided with the instrument (GER, 1999), one can exactly know the
area of sensed region (figure 1). However, we found that this data is not quite right. An
experiment was, therefore, performed to test the real FOV. In this experiment, the sensor was
set up to aim a black board with a shooting distance of 97 cm, and was kept stationary. The
size of the black board is 20 by 20 cm, much larger than the expected FOV area. Then, a 1-cm
wide white strip was moving 1 cm each time from left to right and from top to bottom over the
black board, as sketched in figure 2. Spectral data were collected for each movement. The
overlapped area of the white strip and the FOV area, therefore, can be estimated by calculating
the spectral distance between each measurement and the background spectra. Figure 3(a)
shows the spectral distances of the horizontal movement, and that of the vertical movement are
shown in figure 3(b). Finally, the estimated FOV area and its center position relative to the
laser spot are sketched in figure 4. The shape of the FOV area is obviously not a square. It
could also be true that the GIFOV of most scanners is not a square either. It means that the
fraction coefficients derived from spectral unmixing do not correspond with the proportions of
the þ endmembers distributed within a square pixel. A variation of a spatial distribution of
objects within a pixel would result in a large difference.
Figure 1: The FOV area of the GER 1500 with
a shooting distance of 97 cm.
Figure 2: The FOV test experiment
setup.
(a) the horizontal movement
(b) the vertical movement
Figure 3: The spectral distance between each measurement and the background spectra
measured in the FOV test
Figure 4: the estimated FOV area and its center position relative to the laser spot.
In addition, the boundary of the sensed region is not a clear cut, due to the sensor spatial spread
function (Wu and Schowengerdt, 1993). This could be another factor that would distort a
spectral mixture. To avoid this problem in data collection, the size of target objects should be
smaller than the FOV area.
4. Test Data
A data set was collected to test how the spectral mixture modeling fits the real data. The target
objects are two mixed materials, a soil region as the background and a proportionally increased
vegetated area as the foreground. In order to ensure the proportion of the foreground is linearly
increasing, 8 pie-shape green leaves in the same size were piece-by-piece put into the sensed
region, as shown in figure 5. If the area of a green leaf is x% of the FOV area, the mixed
proportions of the foreground will be 0%, x%, 2x%,…,8x%, and 100%. For each mixture, 10
sample data were collected. The spectral curves of all the sample data are shown in figure 6.
The estimated spectral noises based on the collected data are shown in figure 7. The measures
of spectral distance between each mixture and the pure background in feature space are also
graphically shown in figure 8. The distance increases linearly with respect to the increase of
foreground percentage.
Figure 5: The distribution of the target objects for the data collection.
Figure 6: The spectral curves of the sample data.
Figure 7: Estimated spectral noises.
Figure 8: Spectral distance variation.
