Multi-Object Detection Using Enhanced Computer Vision Models for Underwater Systems

In robotics, computer vision modeling empowers robots to perceive and interact with their environments effectively. This paper presents a groundbreaking effort to develop a computer vision model tailored to underwater robotic systems aimed explicitly at detecting multiple objects of interest beneath the water's surface. The primary focus of our model lies in accurately differentiating between native and zebra mussels, two distinct types of freshwater mollusks with significant ecological implications. Specifically, our model distinguishes native and zebra mussels, two distinct freshwater mollusks. Native mussels are integral to maintaining water quality and serving as essential food sources and habitats for various aquatic organisms. On the other hand, zebra mussels, an invasive species, pose a significant threat due to their rapid colonization and replication abilities in new environments. In this research, we present a mobile application that harnesses the power of this cutting-edge model, allowing it to run efficiently on smartphones. This application can potentially empower citizen scientists in their underwater species detection endeavors. Considering the novelty of this research field, the availability of datasets for training such models remains limited. To contribute to the scientific community, we provide datasets comprising native and invasive zebra mussels, which were employed to train our sophisticated model. We have also presented our preliminary work on identifying fast-moving objects such as zebrafishes. Through this work, we aspire to make meaningful advancements in underwater computer vision and facilitate vital environmental monitoring and conservation efforts.