We applied a deep convolutional neural network (CNN) to recognaize the characterisitic signatures from embedded objects in GPR data. For the application of the deep CNN, we first created a number of categorized images based on a migration procedure. Using the procedure, we showed that the difference without and with a migration can effectively identify GPR images with characteristic reflection events, and the number of categorized images we created was 53510. The application of the deep CNN showed that the total error was less than 0.055. The error was significantly smaller when the same processing was performed by a conventional neural network. We also examined the error when images processed by migration, and found that the training image type does not significantly influence the error. |