Research On Route Measurement Of Self- Propelled Ship Model

To ensure the safety of the navigation,self-propelled ship model test is widely used in navigable administer engineering to visually and really reflects navigable condition,offering reasonable suggestions for the design of route.The route of self-propelled ship model is an important parameter to evaluate the effect of self-propelled ship model test,so it is crucial to accurately and quickly measure the route of self-propelled ship model.At the same time,the rapid development of deep learning has brought new technological innovations to many fields,so this paper introduces deep learning into the field of water transport engineering,and proposes deep learning based route measurement approach of self-propelled ship model.The high-precision Faster RCNN algorithm is used for the initial positioning of self-propelled ship model,and then,the coordinate information of the prow and stern is extracted by frame difference and Freeman chain code.Experimental results show that the Faster RCNN algorithm has a high detection accuracy for selfpropelled ship model detection with average precision of 98.5%.Additionally,to visually display the route of self-propelled ship model in the river model,this paper develops a visualization software,which has a powerful data post-processing function.It can map out the route of self-propelled ship model and the map of river model in real time,achieving rapid measurement of self-propelled ship model's route.This paper is not only a theoretical study of route measurement approach of self-propelled ship model,but also improves the function of the route measurement system of self-propelled ship model from the perspective of productization,so this paper is of theoretical and practical significance.The main work of this paper is as follows:(1)Target detection and target recognition algorithms have been surveyed mainly from motion information based algorithms and statistical learning based algorithms.The basic principles and implementation methods of background subtraction,frame difference,and optical flow are introduced respectively.Besides,the application scenarios,advantages and disadvantages of the three algorithms are summarized.Moreover,this paper introduces the structure of convolutional neural networks and the role of each network layer convolutional layer,pooled layer,and fully connected layer),describes the basic flow of convolutional neural network's training methods.(2)The design scheme for route measurement system of self-propelled ship model is presented by field visit to environment of ship model test.For the reason that river models have different sizes and it is difficult to monitor the entire river model under a video acquisition device,this paper proposes to use multiple video acquisition devices to monitor the entire river model,and this system has scalability.Generally speaking,the video capture devices can be freely expanded according to the size of the measurement range without being limited by the design of the system itself.(3)The video acquisition devices in this paper have free perspective,which through the projection transformation algorithm to obtain aerial view.So it avoids the complicated setup process of video acquisition devices before every measurement and has better robustness to complex terrain and illumination.To improve the detection accuracy of the system,this paper uses high-precision Faster RCNN algorithm to detect the ship model,and the,it introduces the realization principle of Faster RCNN algorithm and the making of data set.In addition,the software development process of this self-propelled ship model route measurement system and general functions of each module are introduced.(4)To verify the correctness of the relevant theories and the feasibility of the proposed design scheme,a series of experiments are carried out.The precision-recall curve and average precision obtained from the network model demonstrate the effectiveness of the Faster RCNN network model for ship model detection.The main steps' results of route measurement approach and the route of self-propelled ship model plotted by the analysis software are shown,which visually shows that the proposed approach achieves accurate measurement of self-propelled ship model route.In general,the proposed deep learning based route measurement approach for self-propelled ship model can be applied to ship model test successfully.