Abstract:To the technical challenge of real-time online monitoring of surface obstruction flow in inland waterway hubs,this paper conducts research on the design of an intelligent recognition system that integrates deep learning and multimodal data augmentation.The aim is to break through the bottleneck of traditional measurement techniques that cannot monitor in real-time online and improve navigation safety and security capabilities.A surface flow intelligent recognition system is designed,which includes modules such as image acquisition and preprocessing,object detection and recognition,flow velocity calculation and flow classification,and data visualization.Three core algorithms are proposed:a large-scale particle image velocimetry algorithm based on deep learning,which calculates the fluid velocity field through LSPIV and optimizes the calculation results using BP neural network,reducing the relative measurement error to 3.48%.A multi-stage risk perception algorithm based on YOLO-BP network,combined with YOLOv5 object detection and BP neural network classification,achieves dynamic assessment of flow obstruction risk with an accuracy rate of 98.34%.A special working condition data augmentation algorithm for interference such as reflection,rain and fog,using wavelet analysis,sparse coding and dark channel theory,effectively improves image recognition quality.Through actual testing and verification of the Three Gorges and Gezhouba hub section of the Yangtze River,the single point monitoring range of the system is about 400 m × 500 m,supporting continuous operation for 7×24 h.It can provide high-precision,all-weather flow monitoring technology support for inland waterway hubs and has reference significance for promoting the development of intelligent shipping technology.
