Underwater Pipeline Detection using Deep Learning Techniques

With the increasing demand for underwater infrastructure, the need for efficient and reliable methods for detecting and monitoring underwater pipelines has become paramount. This study gives a complete assessment of YOLOv8, an object detection model, in connection with detecting underwater pipelines, which can be utilized for transfer learning and future research. A comparison with prior versions, YOLOv5 and YOLOv7, high-lighting the importance of computational speed in a range of scenarios. The study examines the YOLOv8 framework and analyzes image preprocessing methods for improving detection accuracy under a variety of situations, including low light, noise, and background. Furthermore, hyperparameter tuning experiments are carried out to improve model performance by altering batch sizes, epochs, learning rates, and augmentation methods. Mean Average Precision is used to evaluate the model (mAP), which is a commonly used measure to assess how well the model can detect objects. The research evaluates the models’ resilience and generalization capabilities using mAP scores generated across a variety of test situations, highlighting the importance of YOLOv8 in infrastructure maintenance, environmental monitoring, and disaster response.