| In recent years,with the frequent occurrence of chemical building safety accidents,which has aroused high attention from the society,how to strengthen safety management has become an urgent issue.Therefore,in the process of chemical building construction,a large number of drawing reviews for construction drawings are tedious and arduous.The research on the intelligent image review system has certain theoretical and practical significance.This article focuses on the analysis and research of expert signature recognition in the process of image review,and accurately recognizes expert signatures.The intelligent review system for construction drawings can save time for reviewers,shorten the review process,and complete the construction of chemical building more efficiently and quickly.The work done in this article is as follows:First,by intercepting the self-built data set of drawing experts’ handwritten signature part: handwritten signature data set,joint public signature data set Sig Comp2011 for experiments.Aiming at the problem of a small number of signature images,the data set is expanded through image perspective transformation,and the image is enhanced;based on the Open CV tool,combined with image preprocessing technology,the signature image is binarized,normalized,corroded,expanded,and contour extracted,Feature extraction and other operations to remove interference information.Then,starting with the convolutional neural network model,choosing the most suitable network model,and experimenting with the self-built data set and the public data set on the AlexNet,VGGNet and ResNet models.The results show that for the self-built expert signature data set in this paper,the AlexNet model has the best recognition effect,with a recognition accuracy rate of 97.67%,while the VGGNet model has an accuracy rate of less than 1%.Therefore,certain improvements have been made to the VGGNet and AlexNet models respectively.For the VGGNet model,convolutional downsampling is used to replace pooling downsampling,and the Leaky ReLU function is replaced by the ReLU function,and the accuracy rate is 98.29%.For the AlexNet model,convolutional downsampling is used to replace pooled downsampling;FPN network fusion features are used;cascade thinking is used,and three classifiers are used for training,and three classifiers are used for classification and recognition at the same time during prediction,with an accuracy rate of 99.3 %.The experimental results show that the improved model proposed in this paper has a better recognition effect on the self-built expert handwritten signature data set,and it is suitable for the intelligent review data analysis task proposed in this paper.Finally,the research and development of an intelligent review system for construction drawings of chemical plants is mainly aimed at the research of the automatic recognition module of expert signatures,combined with the above research results,the signature recognition is applied to the intelligent review system,which realizes the automatic recognition of expert signatures,and is oriented to the intelligent review system of chemical plant construction drawings.The research has certain theoretical and practical significance. |
