| The function of template replication and creation is provided in the function group of NX platform ship design,which is convenient for the replication and creation of ship plate structure.When using the template copy function to create the transverse bulkheads of complex ships,the problem of creating the transverse bulkheads is easy to occur.It is mainly caused by the large irregular changes of the structural dimensions and intersection angles of the ship deck,which is difficult to be completely solved by using the traditional logic code.As a result,the template copy and creation function can only be used under the condition of regular deck,and transverse bulkheads cannot be created quickly in many cases,which increases the work time of designers.In view of the above problems,this paper proposes to use the image semantic segmentation technology based on deep learning to recognize the transverse bulkhead region created by template copying,and divide the correct region and the wrong region;then,based on the improved image semantic segmentation model,the function of recognition and correction of transverse bulkhead template copying creation region of ship design functional group under NX platform is developed.In this paper,based on NX platform ship design functional group model data set,a variety of mainstream image semantic segmentation models(FCN,U-Net,SegNet,PSPNet,DeepLabv3+,RefineNet,DaNet)are compared.The results show that the segmentation accuracy of PSPNet(Pyramid Scene Parsing Network)is the best,which reaches91.86%mIoU in the data set of this paper.However,the segmentation error will still appear in the small-scale features such as the ship deck intersection area,which may lead to the reduction of the correction performance of the final transverse bulkhead region.Aiming at the problems of PSPNet,this paper proposes an improved PSPNet model RFPSPNet(Recursive Feature Pyramid Scene Parsing Network)based on recursive feature pyramid.Compared with the PSPNet model,using Res Net(Residual Network)as the main feature extraction network has strong multi-scale feature extraction ability.Through experimental comparison,the model achieves 95.03%mIoU in the data set of this paper.Moreover,the segmentation results are more regular,and in the process of small-scale structure segmentation,such as ship deck intersection area,there is basically no large area segmentation error that may eventually lead to the correction error of transverse bulkhead region.Based on RFPSPNet model,this paper develops the function of region recognition and correction of ship transverse bulkhead template replication.The function is mainly composed of model input image capture module,model output data and transverse bulkhead region matching and correction module.Through the existing transverse bulkhead template replication problem model of the ship design functional group under NX platform,the performance of the function is tested.The verification results show that the function module can accurately identify and correct the error creation region of transverse bulkhead,and the function execution time and memory occupation are within the allowable range of NX platform.This function successfully solves the problem of wrong selection of creating region for copying ship transverse bulkhead template,greatly reduces the user’s work time,and improves the efficiency and accuracy of modeling. |
