ADVANCING MARINE ECOSYSTEM CONSERVATION: OBJECT DETECTION WITH AUVS AND REAL-TIME ALGORITHMS

Advancements in computer vision, particularly in the realms of image segmentation and object detection, are pivotal for marine ecosystem monitoring, which is crucial for conservation endeavours. Given the inherent risks involved in human observation of marine environments, the development of Autonomous Underwater Vehicles (AUVs) is imperative for such tasks. In fisheries, especially in aquaculture, employing AUVs for fish monitoring holds significant importance in sustaining marine ecosystems. This research investigates real-time object detection utilising advanced algorithms — specifically, masked convolutional neural networks implemented via Detectron2, a library for detection and segmentation algorithms developed by Meta AI Research. Leveraging the Google Open Images v7 dataset, which encompasses a wide array of fish types differing in size and shape, the study evaluates performance metrics including precision, recall, Intersection over Union (IoU), and mean Average Precision (mAP). Through multiple training iterations under a transfer learning paradigm and building on the LVIS large vocabulary instance segmentation pre-training dataset, the study achieves commendable detection performance, showcasing the effectiveness of the proposed framework for automated, scalable marine ecosystem monitoring