Research Output
Shape Recognition of GPR Images using Hough Transform and PCA plus LDA
Recognition of different utility pipes size of ground penetrating radar images at different penetration depth
1 - 2 of 2
Publication
Ground penetrating radar (GPR) is a nondestructive test used for shallow subsurface investigation such as land mine detection, mapping and locating buried utilities. In practical applications, GPR images could be noisy due to system noise, the heterogeneity of the medium, and mutual wave interactions. Hence, it is a complex task to recognize the hyperbolic pattern from GPR B-scan images. Thus, this project proposes combined shape recognition of buried objects using Hough Transform (HT) and PCA plus LDA in GPR images. The use of HT is justified because it has the property of transforming global curve detection into efficient peak detection in the Hough parameter space. Whereas PCA plus LDA tries to maximize between-class scatter while minimizing within-class scatter. In this framework, the preprocessed GPR images were extracted using HT. The extracted HT features were subjected to PCA plus LDA to map them from high into lower dimensional features. Then, the reduced PCA+LDA features were used as input to the k-NN classifier to recognize four geometrical shapes cubic, disc, and spherical of the buried objects. Based on the results obtained, the average recognition rate of reduced HT features using PCA plus LDA was achieved 85.30% thus shows a promising result.
Publication
Ground Penetrating Radar (GPR) is a geophysical locating method that uses radio waves to capture images below the surface of the ground in a minimally invasive way. It also requires two main essential equipment which is a transmitter and a receiving antenna. To address the problem, this project proposed the hyperbolic recognition of different utility pipes of GPR images at different level of penetration depth. In this framework, the raw data of GPR images were firstly to be pre-processed. The grayscale images were cropped, resized, and enhanced to increase the contrast of the features of the image. Then, the pre-processed GPR images were extracted using the Histogram of Oriented Gradient (HOG) method with three different windows. The extracted HOG features were then used as input to the k-Nearest Neighbor classifier. A series of experiments has been conducted using 10-fold cross-validation technique for training and testing the GPR data. Based on the result obtained, it shows that at depth 20cm the average accuracy is about 99.87%, whereas at depth 40cm the average accuracy achieved 100%. Thus, the result shows that the extracted HOG features exhibit the significant information of hyperbolic signature of different pipe size with different depth of buried object. Therefore the results seem promising in recognizing the hyperbolic of utilities.