objects and
rk, the data
nent) can be
S data. The
ces with the
Fig. 1). The
S data d, the
wy, p) of all
iulti-spectral
users (map
scene, and
») data given
nodel.
probabilities
] in GIS and
Classification
metry given
segmentation
dio, p) was
n p the set of
eferred to as
ob(oyld)]".
given initial
s to generate
ead function.
p.
Ali Akbar Abkar
U2
RS Data
Acquisition
(new) RS data GIS-data and/or GIS
knowledge
(World Model)
Radiometric Geometric ^7
hypothesis generator | *
Hypothesis map
Hypothesis evaluator
likelihood generator
Likelihood map
parameters
L^
“|
ax. likelihoo
per object
hypothesis
Figure 1. Conceptual representation of
likelihood-based segmentation and
classification method. In this Figure,
processes are represented by rectangles and
the obtained information (data sets) are
represented by parallelogram.
Yes
Object/sensor
parameters
a hypothesis evaluator, which evaluates each iteration of the hypothesis generation process calculating the degree
of agreement or disagreement between the evidence map and the hypothesis map. This results in the average
likelihood over all classes and all objects. When multiplied with a cost/benefit matrix a monetary utility function is
produced.
- Ifthe utility is maximum then stop and update the GIS.
- If the utility is less than maximum, then change the model (parameters) and go to step 2 and generate new
geometric hypotheses maps.
A typical feature of the method is the separation of the complex combined probabilities of multi-spectral data, object
class, object geometry, sensors and sensing into a radiometric model and a geometric model. Bayesian inversion of
the radiometric model leads to radiometric evidence maps. The geometric hypotheses maps are generated from
parameterized object models. Geometric models are constrained by fixed objects and they include elements from the
sensor model, such as the point-spread function and the mapping of three-dimensional scenes into two dimensional
sample grids. The utility function consists, for each iteration of geometric hypotheses generation, of the matrix over
the crossing of all evidence vectors and all hypotheses vectors. This "confusion" matrix is multiplied (inner product)
with a benefit/cost matrix. The user of the combined RS&GIS system can optimize the results either in the
likelihood domain (unit cost/benefit matrix) or use several different cost/benefit scenarios. Residual error maps
show the distribution of residual errors. Study of the utility(parameter) function gives the overall quality in the
minimum value of utility.
3. EXPERIMENTATION, VERIFICATION AND EVALUATION OF THE FRAMEWORK
The LBSC is applied to the problem of identifying and estimating the expansion or depletion of forest in the study area
located in Thailand from 1977 to 1989 with the use of satellite remote sensing data, such as Landsat-TM images.
3.1 Study Area and Data
The district of Phrao in Chiang Mai province in the northern region of Thailand was selected as a relevant study area.
The general landscape is hilly and mountainous surrounding a small lowland area where settlements and agriculture are
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 11