943
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bl. Beijing 2008
3.3 Geometry correction result
Based on TM data (adjusted by using 1:100,000 topographic
maps), image geometry correction was made. 24 scenery TM
CCD images (Fig.3) were necessary to carry on the comparison
correction. In the correction process, used neighbour element
resample, constant area division conical projection, and applied
national unified central meridian and the double standard
parallel, chose 397 control points(road junction, rivers junction
and obviously object points). The mountainous area had the
geometry distortion slightly, increased the control dot density.
Because there was not obvious terrain feature in the Bohai Sea
area, no point was chosen. After finishing image correction,
mountainous area error was between 1.5-2.5 elements; plain
area error was in 1-2 elements. Through contrast the same
object point between remote sensing data and the topographic
diagram, computed plane dimension absolute error to confirm
“Beijing 1” Micro-satellite data precision. Image spot position
error was in 2.02%, could satisfy the application request.
Fig.2 "Beijing 1" Micro-satellite spectrum characteristic of
different land types
Fig.3 Contrast image Coverage area between “ Beijing 1 ”
Micro-satellite image and TM
3.4 Classification result
Between forest land and lawn, construction land and sand land,
urban land and countryside residential area, spectral signatures
were similar. So they were merged separately. Finally
classification template included 5 kinds: The reservoir, the pit
pond, the farming, the construction land and sand beach, the
urban land and the countryside residential area, the forest land
and the lawn. After the classification, the classification post
processing was made to obtain the ideal classification effect,
mainly including colour evaluation, filtration analysis, recode
and so on. Then scientific classification result was obtained.
Classification chart was as Fig.4.
3.5 Classification precision appraisal
By computation, classification precision was obtained as table 2:
Classification
Overall
Kappa
method
precision
coefficient
Maximum Likelihood
89.45%
0.8436
Minimum Distance
73.93%
0.6240
Mahalanobis Distance
84.82%
0.7848
Tab. 2 Classification precision contrast in different method
We can see that classification precisions were high in maximum
likelihood and Mahalanobis methods. Classification precision
was low in the minimum range. The partial reasons were
intrinsic flaws of the algorithm. The minimum range method
had not considered mobility of the land use type. For instance,
the urban land image element difference was very big. The
distance was big to the template's mean value. Then, some
elements that belong to the urban land use type were possibly
divided to other categories by mistake. On the contrary,
regarding some types that internal change was small, such as
reservoir and pit pond, possibly classification elements were
excessive, we divided them to the type that did not belonging to
the type.
Because of many kinds of factor limit, for example, phase and
same spectrum different things, three methods all had the
phenomenon that terrain feature was divided by mistake.
Spectral signatures between partial forest land and central city,
reservoir, pit pond were extremely similar, several categories
had wrong classification phenomenon. The countryside
residential area assumed non-rule geometrical shape, and
distributed periphery the farm land. In classification cluster
processing, some land types that were smaller than 2 elements
were merged to neighbour big category. Therefore, individual
countryside residential areas were divided to the farm land type
by mistake.
4. CONCLUSION
The study indicated that “Beijing 1” Micro-satellite CCD data
has the certainly application value in the land use research field.
“Beijing 1” remote sensing image coverage is broad. Wide
range monitor has obvious superiority. The time resolution is
high, on the aspect of land use dynamic monitor, application
potential is huge.
Therefore the data have usability in land use research area, can
be used in land use/cover, resources investigation and research,
become one new data resource which the remote sensing data
renew.
In future, image geometry correction needs to study further,
according to its own characteristic. “Beijing 1” Micro-satellite
data quantity is bigger; the geometry correction process needs
more time-consuming. So When guaranteeing on high
correction precision, it is necessary to guarantee quick
processing, the operating speed, and reduce gamma controller
change as far as possible. So the new request was set about
geometry correction method.