Research And Implementation Of Image-based Wallpaper Defect Detection And Recognition Method

In recent years,the production and use of wallpaper has doubled and redoubled.A large number of products and high-strength production lines make workers work under too much pressures and aging of production equipment,leading to wallpaper productions which has defects.Therefore,defect detection of wallpaper products has become an important part of wallpaper production line.Because the current wallpaper defect detection system is mainly based on hardware monitoring,and requires a lot of manpower to detect,high cost and low detection accuracy,image detection has become the development trend of wallpaper detection.The method of image detection has low cost and high speed.Because of many lines and complex patterns in wallpapers,which makes the actual defect recognition and detection is inaccurate.Therefore,a complete and accurate wallpaper defect detection system plays an important role in wallpaper production,At present,there are four kinds of common wallpaper defects: cracks,holes,folds and black spots.In this paper,experimental research is based on the wallpaper images to detect and recognize four kinds of wallpaper defects,including cracks,holes,folds and black spots.Then design a detection application system and transplant the experimental models into practical application to ensure the accuracy of the algorithm.The main work and innovations of this paper are summarized as follows:(1)In view of the complex background of wallpaper,which affects the defect detection results,according to observe and analyze the original wallpaper images and preprocess them.By comparing the defect images in RGB,HSV and LAB three color spaces,the experiment chooses to process in RGB color space,then superimpose the three color components of R,G and B,in order to enhance the defect,suppress and filter the background image of wallpaper.It is preparing for detection of defects in behind.(2)For defect detection,this paper firstly fits the gray histogram of wallpaper image to a Gaussian function,and uses the parameters of Gaussian function as input value,segmentation threshold as output value,and trains with the generalized regression neural network model to segment the black spot defect of wallpapers.(3)Aiming at the problem that the generalized regression neural network can detect the limited types of defects,an improved Otsu method for defect segmentation is proposed based on the idea of segmentation method.By improving the traditional Otsu method,the segmentation threshold is changed with the complexity of the foreground and background,so that the defect can be segmented.Finally,the location of the defect in the original image is obtained from the segmented binary image.This method resolves the problem of single defect type in generalized regression neural network segmentation,and realizes the segmentation detection of cracks,holes,folds and black spots.(4)Aiming at the problems of defect recognition speed and accuracy,comparing with BP(Back Propagation,BP)neural network,Radial Basis Function(RBF)neural network and BP neural network optimized by genetic algorithm(GA)to select the model which has the fastest convergence speed and the highest recognition accuracy.To classify and recognize the defects after detecting defects,first of all the feature that can represent the defect part is selected.In the experiment,seven feature components of gray and geometric features of defect images are selected as input features of classification model.The normalize input features vectors,so that the numerical range of the calculated features can be unified.Then the weights and thresholds of BP neural network are numerically coded,the fitness is calculated by the expected output and the actual output of the network,and the model is modified by the selection,crossover and mutation of the genetic algorithm to operate the new chromosomes,thus the optimized BP neural network is obtained.Finally,comparing the convergence speed and classification accuracy of BP neural network,RBF neural network and GA-BP neural network.GA-BP neural network is chosen as the classification model to resolve the problems of slow convergence speed and low classification accuracy.(5)After validating the experimental research,developing a complete wallpaper defect software system based on the experiment result.This system can not only complete defect detection,defect recognition and view the results,but also make statistical analysis of the test results.Producing abundant statistical charts and tables for users to observe and maintain production equipment.Experiments show that by comparing to other methods in detecting four kinds of wallpaper defects such as crack,hole,fold and black spot,the improved Otsu method is more accurate,the segmented defects are more complete,and the detection speed is faster because of the low computational complexity of the method.After calculating the features of the image,the feature vector is used as input and GA-BP neural network is used as classification model to classification and recognition.Compared with other models,GA-BP neural network has faster convergence speed and higher classification accuracy,which has an average recognition accuracy of 95.25%.Moreover,this model has high detection and recognition efficiency which recognizing a single image costs 0.923 s.By using the combination of improving Otsu algorithm and GA-BP neural network to recognize the wallpaper defects can be fast and accurate,and the research results provide a reference for wallpaper production quality detection.