Study of Hyperbolic Cavity Pattern Using BEMD in the Presence of Noisy Environments

This study investigates the use of Ground Penetrating Radar (GPR) to detect cavity with presence of buried metal pipes in noise-corrupted environments, such as those containing rocks or bricks, where signal interference complicates detection. The main objectives are to analyze the interactions between hyperbolic cavity patterns and two buried metal pipes, develop a standardized GPR data database for various noisy conditions, and improve detection accuracy through enhanced data processing. A controlled testbed using dry sand as medium was constructed, embedding a simulated cavity, two metal pipes, and concrete blocks to represent real-world scenarios. GPR data were collected using the MALA CX12 system with a 1.2 GHz antenna and processed through grayscale normalization, region of interest cropping, and image resizing. Feature extraction was carried out using texture-based features, namely Gray Level Co-occurrence Matrix (GLCM) and Bidimensional Empirical Mode Decomposition (BEMD). The extracted features were then subjected to classifiers such as k-Nearest Neighbors (k-NN) and Support Vector Machine (SVM) to recognize cavity in the presence of noisy environment. A series of experiments have been conducted on the collected GPR data. The results obtained show that the texture-based features of BEMD using SVM classifier had achieved the highest accuracy of 98.89\%. The result demonstrated that the BEMD has outperformed the GLCM texture-based features. These results highlight the effectiveness of combining GPR with BEMD and machine learning for accurate anomaly detection, offering a promising framework for applications in construction, utility mapping, and underground diagnostics.