Pipeline Localization Using Unsupervised Neural Network Technique

Abstract

The main purpose of this paper Is to identify and detect the position of pipeline on side scan sonar

image. This work is performed in two steps. The first one is to split an image into regions of uniform texture

using the Gray Level Run Length Matrix (GLRLM) which is widely used in texture segmentation application.

The last one addresses the unsupervised leaning method based on the Artificial Neural Networks (Self-Organizing Map or SOM). The result of SOM network based on data histogram visualization is determined as

the comparative model of object of interest. To increase the performance of our method, we propose a penalty

function used for estimating the position of pipeline. After a brief review of both techniques (GLCM and SQM) we present our method and some results from several experiments on the real world data set.