Unveiling Underwater Structures: MobileNet vs. EfficientNet in Sonar Image Detection

Abstract

Deep learning techniques have led to an increased use of Convolutional Neural Networks (CNN) in recognizing images for marine surveys and classifying underwater objects. Applying CNN for the automatic identification of targets in side-scan sonar (SSS) images can boost accuracy and efficiency. Using transfer learning and deep learning as lenses, this work explores the field of underwater item detection. Comparing performance of MobileNet and EfficientNet in performing the task of underwater item classification from sonar images is the central research objective of this study. These two well-known convolutional neural networks were the subject of a thorough investigation to help explain this. In order to fine-tune pre-trained models on a dataset of sonar images, the study approach used the transfer learning technique. We conducted tests on MobileNet and EfficientNet and thoroughly assessed each of their accuracy levels. This study has two main conclusions. First, showing promise as a strong model for underwater object identification, EfficientNet performed exceptionally well in the classification challenge.