Study On Key Algorithms Of Web Diagnosis On-line For High Speed And Big Width Paper Machine

In the process of paper production,the paper defect is the surface defects such as dirty spots,holes,folds,scratches,dust and cracks.The appearance of paper defects will have a negative impact on the subsequent use,especially on aerospace paper,electrolytic capacitor paper,wallpaper base paper and other special paper with high added value in the later period,which will bring huge economic losses,so it is necessary to carry out accurate diagnosis and timely treatment.The web diagnosis technology is to judge whether the paper contains paper defects by collecting the paper image online with industrial camera.If the paper images containing defects,the further classification and recognition are carried out.Generally speaking,the web diagnosis process can be divided into three main stages:paper image acquisition and preprocessing,paper defects online detection and paper defects recognition or classification.With the increase of the speed and width of the paper machine,the possibility of defects on the paper surface and the amount of collected paper image data also increase exponentially,which makes difficulty of the online detection and the accuracy requirements for recognition and classification also increase.However,the existing web diagnosis process has some problems,such as poor anti-interference,low accuracy of paper defect detection,less types of paper defects that can be identified,and poor scalability of classifier and so on.In order to solve the above problems,the studies on the key algorithms of paper images preprocessing,paper defects detection on-line,paper defects recognition and classification in the process of web diagnosis were carried out in this paper.The main work and contribution of this thesis were shown as follows:I.Study on de-noising algorithm in paper image preprocessing based on DCT(Discrete Cosine Transform)homomorphic filteringIn the process of paper images acquisition,the noise was mainly generated by the light source.Aiming at the problem that this kind of noise cannot be filtered directly by the commonly used spatial domain and frequency domain filters,a homomorphic filter based on block DCT transform was proposed in combination with the real-time requirements of the paper defect preprocessing.On the premise of ensuring the filtering effect,the block DCT transform was used to replace the Fourier transform of the traditional homomorphic filter to improve the filtering speed.The experimental results show that the proposed filtering algorithm can effectively remove the illumination interference and enhance the image of paper defect area.The time complexity is far lower than the traditional homomorphic filtering algorithm,which could meet the real-time requirements of the subsequent web diagnosis processing.II.Study on paper defect images segmentation algorithms in paper defects detection process based on two-dimensional threshold.Image segmentation was an important technology in the process of paper defects online detection.Image segmentation separated the image regions containing suspected paper defects according to the characteristics of gray,texture and edge of the paper image.Aiming at the problems of low detection accuracy and slow speed in the current paper defects detection process,two paper defect image segmentation methods were proposed based on two-dimensional threshold.For modern paper machine,the speed was fast and the width is wide.Many defects with different gray value may appear in a paper image.For this kind of paper machine,a fast two-dimensional threshold segmentation algorithm was proposed.First,the difference method was used to solve the problem of multiple paper defects.Second,the two-dimensional threshold was solved as two one-dimensional threshold problems to speed up the process.At the same time,according to the gray distribution characteristics of the paper defects image after the difference,the threshold search range was narrowed,and the selection of the optimal threshold was further accelerated.Experimental results show that the segmentation accuracy of this method is significantly higher than the conventional paper defect detection algorithm which based on one-dimensional threshold segmentation under the premise of ensuring the detection speed.In addition,for some paper production lines which require higher paper quality,a low contrast paper defect detection algorithm based on improved ABC(artificial bee colony)optimization was proposed.First,Gabor filter was used to filter out the interference of paper texture.Then,the ABC algorithm was improved to speed up the optimization.Finally,the two-dimensional segmentation threshold calculated by the improved ABC optimization algorithm was used to segment the paper defect images.The experiment shows that the accuracy of proposed low contrast paper defect detection algorithm has been further improved.III.Study on paper defect classification algorithm based on transfer learningThe process of paper defect classification process was to give the types of paper defects and identify them.Traditional paper defect classification needed feature extraction firstly,and then classifiers were designed according to the extracted features.The classification effect depends on the feature selection,and the scalability was poor.In view of the above problems,convolutional neural network(CNN)was used in classification process of paper defects.Aiming at the over fitting problem in the process of training convolutional neural network classifier caused by the lack of paper defect image data,transfer learning strategy was introduced.First,VGG16 convolution neural network was trained by ImageNet,which is a public image data set,and some convolution layer parameters were fixed.Then,the rest convolution layer parameters were fine-tuned by using the self-built paper defect image database.Finally,the full connection layers were improved to meet the needs of paper defects classification.The experimental results show that the proposed method has fast convergence speed in the training process,no over fitting phenomenon,and the accuracy of paper defect classification is high.?.Study on paper defect recognition algorithm based on deformable convolutionThe paper defects classification based on CNN and transfer learning could solve the scalability problem of traditional paper defects classifier.However,the sizes of input images are strictly required by each CNN.In practical application,the collected images of paper defects must be processed to intercept the outer rectangle of the paper defects area and scale the size according to the required by the classifier.At the same time,the classifier cannot directly identify the location of the paper defect area in the collected paper images.In order to solve the above problems,a paper defect recognition method based on the two-stage image detection algorithm Faster R-CNN and deformable convolution was proposed.Due to the small area and irregular shape of the paper defects,the classification accuracy of Faster R-CNN in the paper defects recognition process is low,and the location is not accurate enough.Aiming at the above problems,two deformable convolutions were added after the traditional convolution layer to extract the characteristics of the paper defects more accurately.And then,the deformable RoI(Region-of-Interest)pooling was used instead of the common RoI pooling,which makes the positioning more accurate.Experiments show that the proposed algorithm has a further improvement in accuracy and scalability compared with theprevious algorithm.To sum up,the theoretical research and experimental exploration around the key algorithms were carried out in the process of web diagnosis,such as paper image denoising algorithm,paper defects segmentation algorithm,paper defects classification and recognition algorithms.And in order to solve the problem of lack of data for CNNs training and testing,the data sets of paper defect images were established by combining actual acquisition with image processing for the use of this paper and subsequent research.The above research results show that the algorithms proposed in this paper can achieve high speed and accuracy in the paper defects detection process,high accuracy and good scalability in the paper defects recognition and classification process,and can effectively solve the key problems encountered in web diagnosis on-line process.The results were of great significance for the independent development of web diagnosis system in paper industry,and further improving the paper production quality and enterprise benefit.