Sub-Bottom Imager (SBI) Sub-Seabed Anomaly Detection using Deep Learning

This study presents a deep learning approach for detecting sub-seabed anomalies—such as UXOs, cables, and boulders—using data from Kraken Robotics’ Sub-Bottom Imager (SBI), a mobile acoustic platform that captures 3-D volumetric images of the seafloor. Synthetic Aperture Sonar (SAS) processing and Inertial Navigation Systems (INS) enable high-resolution imaging. SBI data was exported in SEG-Y format, segmented in MATLAB, and labeled into water, sediment layers, and anomaly classes. A deep learning architecture based on Mishra (2021) , integrating U-Net CNNs, RNNs, and wavelet decomposition, was used to classify geological features. Achieving over 90% accuracy for most classes, the model effectively identified high-intensity acoustic anomalies and demonstrated strong potential for automated sub-seabed interpretation, reducing manual workload and increasing detection consistency.