International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B2. Istanbul 2004
metric information, and satellite imageries, due to their
wealth of spectral information, generate spectral data. The
second category of the data sets is the topographic UGD (1e.
3D digital map vector data).
These data sets provide reference information, whereas the
aerial and the satellite images serve to generate the most
recent information and changes. To facilitate the ACD
operations, sets of pre-processing modules are initially
applied to the input data. These are basically fundamental
radiometric and geometric corrections of aerial and satellite
imageries such as the grey scale filtering, the determination
of the sensor attitude and altitude parameters and registration.
More particularly, as the proposed workflow (Figure 1)
indicates, the change detection process is conducted under:
object identification, object extraction, object recognition and
change detection phases.
w of Proposed Methodology
ors
Workflo
Based on
i | Fuzzy Reasoning
i: &
; cognition
os |
} Based on | Based on
Fuzzy Reasoning | if | Neuro-Fuzzy System |
LIN RR CE
"out
cts | | C-Objects | | A- Objects |
a 2 en 5 i eimi
Figure 1. Workflow of presented method for automatic
object recognition and reconstruction
3. EXPERIMENTS AND RESULTS
The proposed automatic change detection methodology was
tested on a 1:1000 scale digital map and a pan-sharpen
IKONOS scene of the city of Rasht, Iran (Figure 2). The
maps have been produced in 1994 from 1:4000 aerial
photographs by National Cartographic Centre (NCC) of Iran.
490
The satellite imagery was acquired on May 28" 2001. During
these seven years time lapse between the generated digital
map data and the IKONOS image acquisition, considerable
changes have occurred in the city (See Figure 2).
Figure 2. 1:1000 planimetric map of the city of Rasht (a),
corresponding IKONOS Pan-sharpen (b), corresponding
aerial photo (c).
For two object classes of building and tree, the preliminary
membership functions for the structural, textural and spectral
information (STS) components are defined based on the
knowledge of an experienced photogrammetric operator.
Where membership functions can initially not be defined
with sufficient confidence, they are tuned and modified by
the learning potentials of the neuro-fuzzy technique. For the
initial training operations of the system, 200 samples as
learning data set and 50 samples as the checking data set
were selected. The more samples are used the more
comprehensive the membership functions would be defined
and hence more reliability for the recognition process. To test
the adaptation potentials of the recognition process for the
modification of the membership functions, 200 samples were
selected so that a great variety of viewing appearances for the
building and tree object classes are covered. Based on these
preliminary training operations the adjusted membership
functions were determined. To assess the capabilities,
reliability and efficiency of the proposed ACD method a
portion of an urban portion of this area is selected. The
selected area is well suited for first experiments with the
proposed ACD method and shows a significant complexity as
regards the proximity of the objects. The results of our ACD
strategy is presented in Figure 3. The visual inspection of the
obtained results demonstrates the high capability of our
strategy.
mm lm t
fete "puce eng
F^ s
Ré gp» PM 00 Jpn (C