Reliable training of convolutional neural networks for GPR-based buried threat detection using the Adam optimizer and batch normalization

摘要

The ground penetrating radar (GPR) is a remote sensing technology that has been successfully used for detectingburied explosive threats. A large body of published research has focused on developing algorithms that automaticallydetect buried threats using data from GPR sensors. One promising class of algorithms for this purpose is convolutionalneural networks (CNNs), however CNNs suffer from ovefitting due to the limited and variable nature of GPR data.One solution to this problem is to use a validation dataset during training, however this excludes valuable labeled datafrom training. In this work we show that two modern techniques for training CNNs – Batch Normalization and theAdam Optimizer - substantially improve CNN performance and reduce overfitting when applied jointly. We alsoinvestigate and identify useful settings for several important CNN hyperparameters: l2 regularization, Dropout, andthe learning rate schedule. We find that the improved CNN (a baseline CNN, plus all of our improvements)substantially outperforms two competing conventional detection algorithms.