This study introduces a novel approach to the critical task of submarine pipeline or cable (POC) detection by employing GoogleNet for the automatic recognition of side-scan sonar (SSS) images. The traditional interpretation methods, heavily reliant on human interpretation, are replaced with a more reliable deep-learning-based methodology. We explored the enhancement of model accuracy via transfer learning and scrutinized the influence of three distinct pre-training datasets on the model’s performance. The results indicate that GoogleNet facilitated effective identification, with accuracy and precision rates exceeding 90%. Furthermore, pre-training with the ImageNet dataset increased prediction accuracy by about 10% compared to the model without pre-training. The model’s prediction ability was best promoted by pre-training datasets in the following order: Marine-PULSE ≥ ImageNet > SeabedObjects-KLSG. Our study shows that pre-training dataset categories, dataset volume, and data consistency with predicted data are crucial factors affecting pre-training outcomes. These findings set the stage for future research on automatic pipeline detection using deep learning techniques and emphasize the significance of suitable pre-training dataset selection for CNN models.

Side-scan sonar (SSS) image classification is crucial for underwater applications, but the trade-off between the accuracy afforded by high-resolution images and the associated computational cost challenges deployment, particularly on resource-constrained platforms like AUVs. This study systematically investigates and quantifies this accuracy–efficiency trade-off in SSS image classification by varying input resolution. Using two distinct SSS datasets and a resolution-adaptive deep learning strategy employing MobileNetV2 and ResNet variants across six resolutions, we evaluated classification accuracy and computational metrics. Results demonstrate a clear inverse relationship: decreasing resolution significantly reduces computational load and processing times but lowers classification accuracy, with the degradation being more pronounced for the more complex four-class dataset. Notably, model test accuracy did not necessarily increase monotonically with resolution. Importantly, acceptable accuracy levels above 90% or 80% could be maintained at significantly lower resolutions, offering substantial efficiency gains. In conclusion, strategically reducing SSS image resolution based on application-specific accuracy requirements is a viable approach for optimizing computational resources. This work provides a quantitative framework for navigating this trade-off and underscores the need for developing SSS-specific architectures for future advancements.