EVALUATION OF LINEAMENT DETECTION ALGORITHMS
USING MULTI-BAND REMOTE SENSING IMAGES
Jinfei Wang
Associate Professor
Department of Geography, University of Windsor
Windsor, Ontario, Canada
Commission VII, Working Group 4
KEYWORDS: Geology, Algorithms, Landsat, Feature identification algorithms, Geologic lineaments.
ABSTRACT
This paper presents a lineament detection method using multi-band remote sensing images. A Linear
feature Network Detection and Analysis system - LINDA has been applied to multispectral satellite data
such as Landsat TM images in the Canadian shield area of northern Ontario. The proposed technique
involves the following steps:
Transform.
Preprocessing, Edge detection, Multi-band combination and Hough
The results demonstrate that the proposed multi-band lineament detection procedure
achieves better performance in comparison with the existing single-band algorithms.
INTRODUCTION
Lineaments are linear topographical or tonal
features on the terrain representing zones of
structural weakness (Williams, 1983). They
may be recognized on and interpreted from
images and maps. Identification and mapping of
lineaments from satellite images is an important
use of remote sensing data in many geologic
applications. On satellite images, lineaments
usually appear as straight lines or edges, but
frequently a lineament may have gaps in it due to
poor contrast of the lineament with its
surroundings or coverage of surface materials.
In visual interpretation and mapping of
lineaments, geologists use their knowledge and
experience to connect lines and edges which are
collinear and broken into a series of segments. It
is suggested that some of the "rules" that are
used by geologists in their image interpretation
can be applied in automated lineament extraction
from digital imagery. This paper presents a
lineament detection method using multi-band
remote sensing images. A Linear feature
Network Detection and Analysis system - LINDA
(Wang, 1993) has been applied to multispectral
satellite data such as Landsat TM images in the
Canadian shield area of northern Ontario.
750
METHODOLOGY
The proposed technique involves the following
steps:
Preprocessing
For the majority of images, it is important to
apply a filter to the image to smooth it and
remove most of the noise. If this were not done,
problems of isolating individual lineaments would
occur later in the analysis. In this study, a
median filter was applied to the original data.
Multi-band combination
As opposed to most existing lineament detection
methods, which use single-band images only, the
proposed method is designed to operate on
several channels and combine different linear
feature information. Principal Component
Analysis (PCA), band ratioing, and their
combination are applied to the filtered channels.
The resultant channel contains edge information
from multiple original channels.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996
(Ty. OY "Irs 85 °° MN m
nN C*
—