GPR Dataset

GPR B-scans of buried cylindirical objects in soil, using OERAD's Concretto system. YOLO Training Dataset has annotated B-scans of GPR hyperbola bounding boxes, masks and apex points in CVAT XML and YOLO annotation formats. It is composed of annotated original B-scans images stored without any processing and after processing, following a multi-channel approach. This approach concisted in storing different data processing techniques in each image RGB channel. These techniques consisted of a GPR processing pipeline, composed of dewow, mean background removal and band-pass filtering, automatic gain control, normalized cross-correlation reference signal matching and Sobel edge filter. Image augmentation techniques were used in the training and validation subsets, indicated in the file names. All of the augmentations were applied equally to each multi-channel encoding, such that the annotations are valid for all of these. The System Verification dataset is composed of original unaugmented B-scans with ground-truth object position.