| As an important part of map quality control,the defect detection of map printing is of great significance to promote the transformation and upgrading of map production and management system.In order to improve the degree of automation of map printing defect detection and the digital management mode of map production,this paper carried out in-depth research on acquisition equipment,image preprocessing,defect detection and defect classification algorithm of map printing,which based on the machine vision detection system composition,image processing algorithm and machine learning.The main research contents were as follows:1.Improved the hardware scheme of the map printing defect detection system.This paper analyzed the reason why it is difficult to detect the defects of light color area de on the map,designed a lighting scheme suitable for the detection of defects on the map,and highlights the areas of general dyeing to the maximum extent.2.Optimized image preprocessing algorithm applicable to defect detection of map printing.To solve the problem of low contrast between paper and area elements,a gray-scale method based on weight adjustment was designed to improve the contrast between light color area elements and paper.An image enhancement algorithm combining Gamma correction and CLAHE was designed,and an optimal parameter determination method is given through experiments.For the noise in image acquisition,bilateral filtering was selected to remove the noise while preserving the edge.3.Established a classification model based on machine learning.Firstly,the paper designed a multi-classification algorithm of support vector machine,and the experiment is carried out through the data obtained in actual production.The classification module based on Faster R-CNN network was designed to meet the requirements of this paper,and experiments were designed to prove that it has better classification effect in small sample data sets.4.Optimized the functional module of the map defect detection system.Improved hardware management module and image processing module.The training module and defect classification module are reconstructed.The training,updating and classification operation of the classification model based on convolutional neural network were added.A defect information management module is added to build a defect training database for the defect classification model,which can help make full use of the defect detection results,guide the map classification,generate the detection report and update the classification model. |
