Automated system for ship draught measurement with component of intelligent systems

Knowing ship draught is vital for calculating cargo mass onboard of marine vessel. Even small error in its measurement can lead to significant loss of cargo. Although many tools and methods for ship draught reading have been developed, none of them can provide sufficient accuracy. Also, every known method assumes human participation in measurement process. To eliminate the human factor, we propose to develop automated system which will provide sufficient for practical needs accuracy. It includes UAV with digital camera based on it, ship clinometers and computing section. We consider a method, based on processing of draught marks video with neural networks. YOLOv5 convolutional neural network has been used for digits detection, and U-Net convolutional neural network for segmentation of water surface area on each frame. Of note, we look over the concept of ship draught in term of stochastic process and took into account ship heel and trim, as they significantly affect the final result.