Recurrent Neural Network based Underground Object Detection using A-scan Ground Penetrating Radar

Abstract

Ground penetrating radar is a highly effective tool to find objects buried underground. But there are still some limitations that prevent it from being applied to a wider variety of applications. The complexity of processing and the opportunity to collect the signal. This work is applied GPR with AI to reduce the processing steps and displayed your object's features immediately. Based on the time series model in Bidirectional Neural Network as RNN in bidirectional format. Because it possible to predict the position and shape of the object simultaneously. The object was buried under a sandbox with a depth of 10 - 50 centimeters and used two antennas to receive and transmit the signal as a GPR machine in this experiment. Found that the accuracy is at 92.8 % and also measured by F1-score in each target positions as well, when using the results of this work to add features and extend it to work with multiple functions, it will be possible to use the GPR used with Neural Network model can be used in the actual situation.