ISPRS, Vol.34, Part 2W2, “Dynamic and Multi-Dimensional GIS”, Bangkok, May 23-25, 2001
ISPRS, Vol.3-
32
PER-FIELD CLASSIFICATION INTEGRATING VERY FINE SPATIAL
RESOLUTION SATELLITE IMAGERY WITH TOPOGRAPHIC DATA.
Mauro CAPRIOLI, Eufemia TARANTINO
POLYTECHNIC UNIVERSITY OF BARI
Dept, of Highway and Transportation
Via Orabona 4, 70125 BARI - ITALY
Tel.: +39 080 5569387 - Fax: +39 080 5569329
E-mail: caprioli@dvt005.poliba.it; tarantino@dvt005.poliba.it
Keywords: Satellite Imagery - Topographic Data - GIS - Per-field Classification
ABSTRACT
The paper analyses an automated system for classifying land use on a locale scale, integrating Remote Sensing and GIS data, with the
aim of extracting meaningful information in performing spatial analysis in urban contexts.
As a first step, the backgrounds in extraction of contextual information from fine spatial resolution satellite imagery in an integrated
RS/GIS system are considered in detail.
Finally, a per-field classification technique, applied to an IKONOS multispectral image with spatial resolution of 4 m by utilising a digital
topographic map, is described and the accuracy assessment of classification with the nature of problems emerged in the procedure is
clearly identified.
1. INTRODUCTION
The information contained in digital imagery, acquired by
Remote Sensing technology, can be used for mapping,
monitoring and assessing the properties of the environmental
and territorial feature elements. Of all these three main
application fields in Remote Sensing, making thematic
cartography by means of automatic classification methods is
surely among the most widespread and in many cases is an
essential preliminary step for further applications.
Up to now spatial resolution of data given by earth observation
satellites has proved inadequate in providing detailed
topographic peculiarities, in specific application domains of
analysis and monitoring of urban environment. Modality of
terrestrial phenomenon representations, augmented geometric
accuracy, temporal flexibility of acquiring, spatial land cover and
appropriate use in spatial modelling terms suggest, instead, the
diffusion and continual use of new very fine spatial resolution
satellite sensors imagery, as data sources for spatial analysis in
urban contexts.
The high level of investigations enabled by interacting with other
disciplinary sectors, can support planners’ activities better, as
Mesev, Longley and Batty (1996) have argued: “our concern
with land use revolves around the central 'urban/non urban'
dichotomy as manifest through physical form, although in
practice finer disaggregations as well as measures of the
intensity of human activities are also desirable".
The informational classes of a thematic mapping are not directly
registered, but must be derived indirectly by using evidence
contained in the spectral data of an image. When we apply
standard procedures of per-pixel multispectral classification the
increase of spatial resolution leads to augmentation in ambiguity
in the statistical definition of land cover classes and a decrease
of accuracy in automatic identification. This problem may be
overcome by means of per-field classification techniques which
involve analysing groups of pixels within land cover parcels.
Such technique, based on the integration of remotely sensed
imagery and digital vector data, has been used to generate land
cover and land use information for more than a decade
(Carbone, Narumalani and King, 1996; Ehlers, Greenlee, Smith
and Star, 1991; Hinton, 1997;). Innovation and power of recent
GIS platforms and analytic flexibility of Image Processing
softwares make the integration of satellite data with numerical
and scaled topographic data much more feasible, and this can
lead to an increase in accuracy of the classification compared
with the per-pixel technique, as well as improvements in
interpretations of results with incorporating spatial variability and
texture inherent in fine spatial resolution imagery.
This work investigates a per-field classification methodology,
applied to an IKONOS multispectral image with spatial resolution
of 4 m by utilising a digital large scale topographic map as a
representative reference land cover, and assesses the accuracy
of classification by comparing the results carried out for both the
per - pixel and the per - field techniques. Finally, the nature of
problems emerged with both procedures is clearly identified.
2. THE EXTRACTION OF CONTEXTUAL INFORMATION
FROM FINE SPATIAL RESOLUTIONSATELLITE IMAGERY
IN AN INTEGRATED RS/GIS SYSTEM.
The accuracy with which land use has been mapped up to now
from satellite sensor imagery from local to national scales has
been limited by the relatively coarse spatial resolution of
instruments. For example, for the land cover maps generated
using Landsat Thematic Mapper (TM) imagery, with a spatial
resolution of 30 m, a considerable amount of detail in the scene
is obscured from the image. The availability of recent
multispectral imagery with very fine spatial resolution has
increased our ability to map land use in geometric detail and
accuracy Aplin, Atkinson and Curran, 1997) for local and
national scale investigations.
However, these sources of imagery are likely to generate other
problems. Even if the radiometric resolution is enhanced (11 bit
for IKONOS imagery), spectral capabilities are generally limited
compared to those of the previous generation sensors (seven
bands for the Landsat TM). Moreover, associated with an
increase in spatial resolution there is, usually, an increase in
variability within land parcels ('noise' in the image) generating a
decrease in accuracy of land use classification on a per-pixel
basis (Townshend, 1992).
Traditional automated classification techniques classify land use
on a basis of spectral distribution of the pixels within an image,
whereby each pixel is associated with the most similar spectral
class. This general method can produce results that are 'noisy'
due to the high spatial frequency of the land covers.
The alternative technique of per-field classification (so called
because fields, as opposed to pixels, are classified as
independent units) takes into account the spectral and spatial
properties of the imagery, the size and shape of the fields and
the land cover classes chosen.
In fact, this approach requires a priori information about the
boundaries of objects in the image, for examples, roads fields. If
the boundaries of these fields are digitised and registered to the
image, then some property of the pixel lying within boundaries of
the field can be used to characterise that fields. For instance,
the means and standard deviations in the four IKONOS bands of
pixel lying within roads fields could be used as features defining
the spectral reflectance properties of fields. Normally, the use of
map and image data would take place within a geographical
information system (GIS), which provides facilities for
manipulating digitised boundary lines (for example, checking the
set of line to eliminate duplicated boundaries, ensuring that lines
'snap on’ to nodes, and identifying illogical lines that end
unexpectedly).
All these (
spatial resc
lead to an ii
for relative
Westmorelc
However, tl
with field b
be carried c
before
during
after cl
Many exan
two. Westr
stage, cart<
classificatio
Alternative!;
basis befor
data for per
The ways w
integrated ir
a) separa
systerr
b) two so
shared
c) a singl
In this studv
of the high
and the pos
spatial infori
3. TH
This researi
per 5 Km) v
(Brindisi) in
