An Unsupervised Artificial Neural Network Method for Satellite Image Segmentation

Mohamad Awad
National Council for Scientific Research, Center for Remote Sensing, Lebanon

Abstract: Image segmentation is an essential step in image processing. The goal of segmentation is to simplify and/or to change the representation of an image into a form easier to analyze. Many image segmentation methods are available but most of these methods are not suitable for satellite images and they require a priori knowledge. In order to overcome these obstacles, a new satellite image segmentation method is developed using an unsupervised artificial neural network method called Kohonen's self-organizing map and a threshold technique. Self-organizing map is used to organize pixels according to grey level values of multiple bands into groups then a threshold technique is used to cluster the image into disjoint regions, this new method is called TSOM. Experiments performed on two different satellite images confirm the stability, homogeneity, and the efficiency (speed wise) of TSOM method with comparison to the iterative self-organizing data analysis method. The stability and homogeneity of both methods are determined using a procedure selected from the functional model.

Received August 14, 2008; accepted September 25, 2008