Picking rebars manually in the data from ground penetrating radar (GPR) surveys of concrete bridge decks is time consuming and labor intensive. This paper presents an automated rebar localization and detection algorithm for performing this task. The proposed methodology is based on the integration of conventional image processing techniques and deep convolutional neural networks (CNN). In the first step, the image processing methods, such as the migration, normalized cross correlation and thresholding, are used to localize pixels containing potential rebar peaks. In the second step, windowed images surrounding the potential pixels are first extracted from the raw GPR scans involved in the first step. Those are then classified by a trained CNN. In the process, likely true rebar peaks are recognized and retained, whereas likely false positive detections are discarded. The implementation of the proposed system in the analysis of GPR data for twenty-six bridge decks has shown excellent performance. In all cases, the accuracy of the proposed system has been greater than 95.75%. The overall accuracy for the entire deck library was found to be 99.60% ± 0.85%.
