In many geophysical surveys there is a pre-defined goal - to detect and locate very specific anomalies, ones which correspond to buried objects (targets). The types of targets range from various types of pipes (metallic or not), to re-bars or wires in walls to land mines. This work presents a novel unsupervised method for automatically detecting targets, and extracting information about them and the medium in which they reside. It does so by efficiently analyzing strips of the B-Scan, and detecting the geometrical signature of a target in the image.
Most existing detection methods are supervised, which means that one has to provide a training set (which can be labor expensive) in order to train a classifier. By contrast, the method presented here is unsupervised and is model based, which alleviates the need to manually annotate a training set.
Another drawback of many existing methods is the underlying assumption of a homogeneous medium. This assumption is greatly relaxed for this method, since it assumes no prior knowledge of the medium. Instead, it learns the mediums properties from the targets themselves. Furthermore, this method is designed to be computationally efficient, applicable in real time applications. The current work presents two version of this algorithm. The first version was designed to detect locally isolated targets (i.e. - without having cross targets interference in the B-Scan). The second version generalizes the first, and is able to locate targets in complex scenarios, at the cost of increasing computational complexity. Both versions were implemented on a commercial GPR system (GSSI's StructureScan Mini XT system) and were tested using multiple systems on real life scenarios. The algorithm was designed to have an extremely high probability of detection - above 95% for the first version, and 98% for the second. The processing time, however, was increased from 20 to 600 microseconds per scan. The experimental results show that both methods are able to detect the targets with high positioning accuracy and sufficiently low false detection rate.
- Prof. Jennifer Dy (Advisor)
- Prof. Octavia Camps
- Prof. Carey Rappaport