A Comprehensive Review of Research Obstacles, Recent Breakthroughs in the Field of Deep Learning-Driven Detection of Marine Objects in Underwater Environments

Abstract

In recent times, the field of underwater marine object detection has gained significant prominence as an indispensable technique in the domain of marine science and engineering. Its potential for ocean exploration has garnered significant attention and practical applications, ranging from monitoring underwater ecosystems to resource exploration and commercial fisheries management. Nevertheless, conventional methodologies encounter formidable hurdles when applied in the underwater environment. This is primarily due to the intricate nature of underwater surroundings, the distinct attributes of marine objects, and the constraints imposed by exploration equipment. These formidable challenges frequently lead to compromised detection performance in terms of speed, accuracy, and overall robustness. Deep learning has risen as a revolutionary catalyst across a multitude of sectors, marine engineering included. In this context, we offer a thorough examination of deep learning-driven methodologies for the detection of underwater marine objects. While a range of sensors can be utilized for this endeavor, such as acoustic sonar, our paper's primary emphasis is on vision-based object detection, owing to its myriad advantages.To provide a comprehensive understanding of this field, we have categorized the research challenges in vision-based underwater object detection into four primary areas: image quality degradation, small object detection, poor generalization, and real-time detection. Our aim is to conduct a thorough examination of recent advancements in underwater marine object detection, illuminating both the strengths and weaknesses of existing solutions for each of these challenges. Furthermore, we strive to enhance the accessibility and usability of the most widely used datasets in this domain through meticulous evaluation. Additionally, we will conduct comparative analyses with prior reviews, particularly those that leverage artificial intelligence techniques, and engage in discussions regarding future trends and developments in this rapidly evolving field.