Pipelines Location From GPR Images Based on Skeleton Extraction and Improved Hash Algorithm

Traditionally, the identification and interpretation of ground-penetrating radar (GPR) profiles for underground pipelines have heavily relied on manual operation, which is subjective and largely dependent on the experience of operators. Conversely, the hash algorithm has been extensively researched and utilized in computer vision to classify and retrieve digital images. In this letter, we introduce a hashing-based algorithm for automatically and accurately identifying pipeline diffractions in GPR profiles. The pipeline identification of the GPR profile is divided into two cases: single pipeline identification and multi-pipeline identification. In the case of single pipeline identification, we proposed an improved hash algorithm based on DC removal to improve the ability to identify hyperbolic features in GPR profiles. However, for multi-pipeline identification, we have addressed the challenge of varying grayscale changes of hyperbolic structures in GPR profiles caused by different materials, which may lead to inaccurate identification. To overcome this, we proposed the skeleton extraction method, which focuses solely on obtaining the shape of hyperbolic structures. By combining these methods, the fingerprints of typical pipeline images are calculated. Subsequently, the measured GPR profile can be scanned using specific fingerprints in the dataset, enabling automatic and accurate pipeline identification. Experimental results have validated the effectiveness of the proposed method.