Motion Error Influence on Segmentation and Classification Performance in SAS-Based Automatic Mine Countermeasures

Abstract

Synthetic aperture sonar (SAS) systems provide capabilities to construct high-resolution images of the seafloor and are, therefore, employed by state-of-the art systems for automatic detection and classification (ADAC) in mine countermeasurement applications. Typically, ADAC systems assume operation on well-focused SAS images. However, in practice, residual motion errors or other errors, e.g., a mismatch in the wave propagation speed, may still be present, leading to a degradation in the quality of SAS images. Consequently, it is of major interest to study the detection and classification behavior of an automatic system under the influence of residual motion and phase errors. First, we train our ADAC system using a database of well-focused SAS images. Subsequently, we use real sonar measurements from different sea trials to build a test database of SAS images where motion errors and sound-speed mismatches are artificially induced into the image reconstruction process to study the impact on segmentation, feature extraction, and classification rate. A relation between the image degradation and the individual tasks of the ADAC system is empirically demonstrated by assessing the image quality using a full-reference method. The obtained results illustrate a severe dependency of the ADAC system performance on the SAS image quality. At the same time, the results highlight the need for both image quality assessment schemes and robust segmentation and feature selection techniques, to improve the reliability of SAS-based target recognition systems under difficult conditions.