16-01-2013, 10:45 AM
THE APPLICATION OF NEURAL NETWORKS, IMAGE PROCESSING AND CAD-BASED ENVIRONMENTS FACILITIES IN AUTOMATIC ROAD EXTRACTION AND VECTORIZATION FROM HIGH RESOLUTION SATELLITE IMAGES
1THE APPLICATION OF NEURAL.pdf (Size: 520.45 KB / Downloads: 63)
ABSTRACT:
In this article a new procedure that was designed to extract road centerline from high resolution satellite images, is presented. The results (road Networks) are fully structured in vector formed in Computer Aided Design (CAD) based system that could be used in Geographical Information System (GIS) with minimum edit. The designed procedure is the combination of image processing algorithms and exploiting CAD-based facilities. In the first step, artificial neural networks are used to discriminate between road and non-road pixels. Then road centerlines are extracted using image processing algorithms such as morphological operators, and a road raster map is produced. Some cleaning algorithms were designed to reduce the existing noises and improve the obtained results. Finally, the edited raster map was vectorized using the CAD-based facilities. Obtained results showed that the structured vector based road centerlines are confirming when compared with road network in the reference map.
INTRODUCTION
Satellite and aerial images are the most important available data sources for map generation and updating of available maps. They provide accurate, easily accessible and reliable spatial information for Geographical Information Systems (GIS). The traditional manual methods of data capture from these images are expensive, laborious and time consuming and do not let full exploitation of available data in image archives. Nowadays when satellite images have highly improved in terms of spatial, spectral and temporal resolutions and Geomatics communities are overwhelmed by the sheer volume of collected images, the necessity of automation in feature extraction and map updating seems urgent.
Roads as one of the most important man made objects are in high concern to be extracted (semi)automatically and many researches have been carried out in this area. Geometrically constrained template matching (Gruen et al., 1995; Vosselman and Knecht, 1995), active contours or snakes (Neuenschewander et al., 1995; Trinder and Li, 1995; Gruen and Li, 1997) and fuzzy set and morphological operators (Mohammadzadeh et al., 2006) are some of the semi-automatic methods for road extraction.
Network Structure
In order to use neural networks in road detection, input layer is
consisted of neurons the same number as input parameters and
output layer is made up of just one neuron that shows whether
the input parameters can represent a road pixel or not. Usually,
one hidden layer is sufficient, although the number of neurons
in the hidden layer is often not readily determined (Richard,
1993).
In this research, a back propagation neural network with one
hidden layer is implemented that uses 500 road and 500
background pixels as its training set.
An adaptive strategy is used to avoid trail and error learning
rate and momentum assignment. In this method, both
parameters are adjusted downwards as half after some training
intervals if the overall training error has increased and upward
1.2 times if the overall error has decreased (Heerman and
Khazeinie, 1992).
CONCLUSIONS AND RECOMMENDATIONS
Image processing techniques such as pattern recognition and
object detection algorithms do not provide fully structured data
for GIS environments. This is mainly due to the fact that the
results of image processing algorithms are raster maps while
GIS environments need structured vector based data as their
input entities. Furthermore, it should be noticed that CAD based
facilities could not be performed directly on source imaged
because of the complicated nature of them, especially when
dealing with high resolution ones.