Algorithm Research And Implementation Of Paper Surface Quality Detection System Based On CNN

As the automation in the paper industry continues to increase,the speed of paper machines is getting faster and faster,and the width of finished paper continues to widen.During the production process,various surface defects of the paper are unavoidable,which is also called paper defect.The existence of paper defects seriously affects the quality of paper and the production efficiency of paper products.If some serious periodic paper defects are not detected and processed in time,a large amount of material may be wasted.Therefore,paper defect detection has become an indispensable important link in the paper production line.Traditional paper defect detection relies on manual detection.With the continuous development of paper production automation,it has gradually been replaced by the methods of machine vision detection.There are many paper defect detection methods based on machine vision,but the detection effect in the actual production process are not ideal.The reason is that there are many types of paper defects,and there are large visual differences between similar paper defects.The paper defect feature quantities corresponding to them are scattered in different feature spaces,and it is difficult to accurately extract all the paper defect features using a unified low-dimensional standard feature quantity.In addition,for difficult paper defects such as wrinkles and scratches,the feature quantities are not significant in each feature space,so that the existing methods have not been effective in identifying these paper defects.Therefore,this paper proposes a paper defect classification algorithm based on CNN.It uses the superior image processing capabilities of CNN to input paper pathogen images,and automatically extracts the deep features of paper defect images through continuous convolution operations and completes the classification work,effectively solving the traditional detection method.It is difficult to determine the characteristics of paper defect and the difficulty of extracting the characteristics of paper defect.This paper completes the paper surface quality detection system research according to the paper surface quality inspection requirements.The main research contents and research results are as follows:(1)Construction of paper surface quality detection hardware system.Based on machine vision technology,a hardware system is built using the structural model of "CCD camera+FPGA+computer".Design and select the hardware equipment used,including camera and lens selection,light source system design,computer selection,and motion control module design and selection.At the same time,complete the design of the FPGA image acquisition board,including the selection of the FPGA and the design and implementation of the main modules of the board(image acquisition control module,image storage module,Ethernet transmission module).(2)Research and implementation of paper surface quality detection system algorithm.Based on the CNN-based paper defect classification algorithm,the research on the algorithm of this system is carried out,including the paper defect detection algorithm research and the paper defect classification algorithm research.The paper defect detection algorithm is completed on the FPGA side.After pre-processing the paper web image,the paper defect area is detected and extracted.Paper defect classification algorithm research based on CNN mainly builds paper defect classification CNN network by studying the structure of convolutional neural network,and uses paper defect samples to complete the training of the network.The trained network model is used to complete the identification of paper defects,and finally complete the task of inspecting the surface quality of the paper.(3)Design and implementation of paper surface quality inspection software system.According to the system testing requirements,design the software system framework.And on the VS2013 development platform based on MFC to design the software technology architecture and software system modules,including software system interface design,data acquisition module design,image module design and database module design.Finally,real-time monitoring of the paper surface quality is completed,and real-time identification,statistics,and storage of paper defects are provided,while providing historical record query functions.Through the overall test of the paper surface quality detection system in this paper to verify the performance of the system,the results show that the hardware system structure designed in this paper can basically meet the detection requirements and has a certain reference for similar visual detection systems,and the software system can run stably.And complete the paper defect identification,statistics and records.The paper defect classification algorithm based on CNN proposed in this paper not only improves the detection accuracy but also has excellent scalability.The experimental results of the algorithm show that the trained model of the CNN network can accurately identify the six types of paper defects such as edge cracks,black spots,holes,folds,bright spots,and scratches.The defect recognition rate is greater than 99.3%,and it only takes 4.675ms to identify a defect image in GPU mode,which fully meets the requirements of paper surface quality inspection.