A Methodology for Assessing Data Augmentation Effectiveness for Target Classification in SAR Images

This work considers a methodology for evaluating data augmentation effectiveness for target classification in synthetic aperture radar (SAR) images. For the analysis, it is considered a dataset of oil platform SAR medium-resolution images. The training dataset of two classification algorithms, namely logistic regression and neural network, was gradually increased with synthetic SAR images. In some cases, the results revealed rising trends in average accuracy as the number of synthetic images increased. By using 50% synthetic images with the application of the data augmentation technique, we achieved the highest average accuracy of the neural network algorithm, reaching 95%, with a standard deviation of less than 5%.