The use of cyber-physical systems currently plays a key role in the development of aquaculture and monitoring the state of water resources. Such systems provide comprehensive monitoring of environmental parameters, the behavior of aquatic organisms and the state of the underwater infrastructure. Pipeline communications are of particular importance, as their damage can negatively affect the ecosystem and operational reliability. The paper proposes an approach to recognizing underwater pipes and determining their orientation on a video stream using a YOLOv9 segmenting neural network and an algorithm for constructing a central trajectory inside a contour. The solution is resistant to perspective and scale distortions, and correctly determines the longitudinal axis of the pipe even when it bends. Based on the dedicated axis, the orientation of the underwater vehicle along the pipeline is ensured, which facilitates navigation and automates inspection processes. To verify the correctness of the algorithm, a modeling complex has been developed that confirms its stability and applicability in monitoring water areas.