Research On Surface Defect Detection And Patching Strategy Of Particleboard Based On Machine Vision

In the production process of particleboard,the detection and patching of surface defects are crucial factors that ensure the overall quality of the final product.Surface impurities,pits,or stains can result in uneven areas,bubbles,or even cracks during the subsequent veneering process,which could negatively affect the quality of the finished boards and other wood products such as furniture.Therefore,achieving accurate and efficient detection and targeted patching of particleboard surface defects on the production line is of both scientific and economic importance.This research topic can provide essential quality assurance for the subsequent veneer and finishing processes of particleboard.This dissertation aims to achieve rapid and effective detection of surface defects on particleboard and to accomplish the identification,grading,and patching of surface defects on particleboard.The study focuses on researching the problems of fast denoising of particleboard images,targeted detection,identification,and patching of surface defects based on machine vision technology.The main research work of this dissertation is summarized as follows.(1)To provide prior knowledge for the construction of data sample set and the analysis of full-text problems,the basic characteristics of common particleboard surface defect types are summarized and analyzed To ensure that the simulated experiments are consistent with actual production conditions,a new sample set of particle board surface defect image data is built by collecting empirical knowledge and sample data from actual production lines.This approach ensures the reliability of the experimental data used in the dissertation.(2)To meet the high real-time requirements of actual production lines in detecting particle board surface defects,the dissertation conducts noise removal experiments on noisy particle board surface defect images in both RGB and YUV color spaces.The characteristics of particle board surface images,such as large amounts of detail information,mixed with Gaussian noise,and single image color,are considered.Based on the further analysis of experimental results and the information distribution characteristics of the particle board surface image in the YUV color space,the dissertation proposes an improved adaptive denoising algorithm.The algorithm performs adaptive denoising on the Y and U image components to preserve image details,while a simplified mean value denoising is applied to the V component using mean filtering to improve computational efficiency.Finally,the dissertation conducts a simulation denoising comparison experiment on particle board surface images with different surface features to verify the rapidity and effectiveness of the improved algorithm.(3)To address the issue of inherent texture interference in particleboard surface images that affects detection accuracy,the dissertation proposes a modified KCF algorithm based on a composite discriminator(Cd-KCF).The improved algorithm aims to ensure detection speed while enhancing detection accuracy.The algorithm optimizes detection efficiency by using a composite discriminator model design scheme,which utilizes a grayscale discriminant classifier in parallel with a chroma discriminant classifier to optimize the utilization of image features.Furthermore,the algorithm improves detection accuracy by adding an FB error correction scheme.The dissertation verifies the accuracy,real-time performance,and applicability of the proposed improved algorithm through comparative experiments on target detection of various defect image sequences on the surface of particleboard.(4)To address the issues of low classification accuracy and overfitting caused by the limited number of samples,the dissertation proposes a Capsule Network algorithm based on an improved CBAM attention model(CBAM-SN).The improved algorithm introduces a CBAM attention model improved by the GELUs function into the convolution layer of the Capsule Network to optimize the feature map of surface defects being detected.This improves the training efficiency and stability of the model.Additionally,the dissertation proposes a defect patching strategy based on fuzzy pattern recognition to achieve suitable patching schemes for matching different surface defects.The strategy includes a defect feature extraction scheme with five parameters and fuzzy pattern recognition to judge the level of surface defects.The dissertation conducts grade judgment experiments on defect sample images using four fuzzy pattern recognition judgment,which indicators to verify the effectiveness of the proposed repair strategy.The dissertation also proposes an improved FMM algorithm to patch surface defects based on the judgment results of defect level.The improved FMM algorithm constructs a new weight function by introducing color factors and confidence factors to make the repair model more stable while ensuring computational efficiency.The dissertation aims to complete the characteristics of high image resolution and high real-time requirements for algorithms in the patching process of surface defect areas.The above researches focus on machine vision technologies,including image denoising,target detection,image recognition,image restoration,and image matching.Existing algorithms were improved by considering the surface image characteristics of particleboard and the production line issues.The study provides a practical approach for detecting and patching surface defects in particleboard.