Speaker
Description
Advancements in computer vision, particularly in image segmentation and object detection, play
a pivotal role in marine ecosystem monitoring—an essential component of conservation efforts.
However, traditional underwater object detection systems often suffer from limitations such as
poor visibility, low-quality imagery, high computational costs, and inadequate performance in
real-time scenarios, especially when faced with diverse marine species and complex underwater
environments. Additionally, the inherent risks and impracticality of manual human observation
in these environments underscore the need for efficient automation. To address these
challenges, the development and deployment of Autonomous Underwater Vehicles (AUVs) for
fish monitoring in aquaculture and fisheries management have become imperative. This
research focuses on improving real-time underwater object detection using advanced
algorithms, specifically masked convolutional neural networks implemented via Detectron2, a
state-of-the-art library developed by Meta AI Research. Utilizing the Google Open Fish dataset—
which contains a wide variety of fish species differing in size, shape, and appearance—the study
assesses performance using metrics such as precision, recall, Intersection over Union (IoU), and
mean Average Precision (mAP). Through multiple training iterations and fine-tuning, the
approach demonstrates significant improvements in detection accuracy, thereby validating its
effectiveness for practical deployment in marine conservation and aquaculture applications.
Keywords: Marine Ecosystem Conservation, Autonomous Underwater Vehicles (AUVs), Realtime Object Detection, Detectron2, Fish Monitoring and Identification
