This paper presents an empirical study that evaluates the impact of different types of augmentations on the performance of Deep Learning (DL) models for detecting imaging artifacts in Synthetic Aperture Sonar (SAS) imagery. Despite the popularity of using DL in the SAS community, the impact of augmentations that violate the geometry and physics of SAS has not been fully explored. To address this gap, we developed a unique dataset for detecting imaging artifacts in SAS imagery with DL and trained a Bayesian neural network with a ResNet architecture using widely used augmentations in DL for computer vision, as well as common augmentations used in the SAS literature. The study shows that augmentations that violate the geometry and imaging physics of SAS can negatively impact supervised classification, but can sometimes improve performance. Overall, the study provides important insights into the impact of different types of augmentations on the performance of DL models in SAS applications.

A test for isotropy of images modeled as stationary or intrinsically stationary random fields on a lattice is developed. The test is based on the wavelet theory, and can operate on the horizontal and vertical scale of choice, or on any combination of scales. Scale is introduced through the wavelet variances (sometimes called as the wavelet power spectrum), which decompose the variance over different horizontal and vertical spatial scales. The method is more general than existing tests for isotropy, since it handles intrinsically stationary random fields as well as second-order stationary fields. The performance of the method is demonstrated on samples from different random fields, and compared with three existing methods. It is competitive with or outperforms existing methods since it consistently rejects close to the nominal level for isotropic fields while having a rejection rate for anisotropic fields comparable with the existing methods in the stationary case, and superior in the intrinsic case. As practical examples, paper density images of handsheets and mammogram images are analyzed.