Underwater Object Detection for Smooth and Autonomous Operations of Naval Missions: A Pilot Dataset

Underwater object detection is essential for ensuring autonomous naval operations. However, this task is challenging due to the complexities of underwater environments that often degrade image quality, thereby hampering the performance of detection and classification systems. On the other hand, the absence of a readily available dataset complicates the development and evaluation of underwater object detection approaches, particularly for deep learning approaches. To address this bottleneck, we have created a new dataset, called National Subsea Centre Underwater Images (NSCUI). It is comprised of 243 images, divided into three subsets that are captured in bright, low-light, and dark environments, respectively. To validate the utility of this dataset, we implemented three popular deep learning models in our experiments. We believe that the annotated NSCUI will significantly advance the development of underwater object detection through the application of deep learning techniques.