Comparison of Spatial- and Frequency-Domain-Based SAR Target Classification Performance

This study compares the classification performance of spatial- and frequency-domain-based synthetic aperture radar (SAR) images for target classification. In general, the image in the frequency domain corresponds one-to-one to the image in the spatial domain through Fourier transform; thus, it can be utilized for target classification because the unique information of the target is preserved in the frequency component. A convolutional neural network model was used for SAR target classification, and the moving and stationary target acquisition and Recognition dataset was utilized. In the experiments using a sufficient amount of training data, there was no significant difference in the performance of SAR target classification based on the spatial and frequency domains. However, when a limited amount of training data was used, the classification accuracy of the frequency dataset was higher than that of the spatial-domain dataset. Thus, effective classification performance was shown not only in the spatial domain but also in the frequency domain for SAR target classification.