Research On On-line Defects Detection System For Edge-sealing Of Veneer Particleboard Based On Machine Vision

Veneer panel is the most popular finished panel in the custom home furnishing industry,accounting for more than 70%.In the production process of veneer panels,the quality of the edge-sealing of the panels directly affects the quality of the finished panels.Nowadays,most of the quality inspection tasks after board edge sealing are completed by visual inspection by quality inspectors.However,the detection speed of this method is slow,the accuracy is low,and the result is unstable.This paper is devoted to researching a set of detection system for sheet edge banding defects based on machine vision to achieve high defect detection accuracy and stable output of detection results.At first,this article analyzes the components of the machine vision system and the mainstream algorithms in defect detection,and studies the usage scenarios of different equipment and algorithms.According to the actual requirements of the subject,a contact image sensor is used to design an image acquisition system,and a deep learning target detection algorithm is selected as the detection algorithm for sheet defects.According to the characteristics of the board and the technical indicators of the system,this paper uses a contact image sensor combined with a mechanical motion experiment platform to build a visual imaging system.This paper studies the imaging characteristics of the contact image sensor,and designs the installation plan of the imaging system according to its characteristics.And design a set of upper computer program that integrates image acquisition function and defect detection function to control the parameters in the process of image acquisition and defect detection.On the experimental platform,the imaging system can collect high-quality images with high contrast and uniform brightness.By analyzing the defect characteristics of the data set,the defects are classified.The structure and function of the deep learning target detection network are analyzed in detail,and the composition of its loss function is studied.The network is trained with methods such as data set processing,pre-training model and gradient descent method.Finally,according to the characteristics of the deep learning network and the actual situation of the data set in this article,several methods of improving the deep learning network are proposed,and a control experiment is set to verify the effect of each experimental condition.In offline algorithm verification,the deep learning network can achieve better detection performance,and the accuracy and recall rate of the model trained under multiple experimental conditions have reached more than 90%.Finally,this article combines the imaging system and the detection algorithm for online defect detection.Experiments have proved that the visual defect detection system developed in this paper can adapt to the 30m/min operating speed of the automated production line for online defect detection,runs stably on the experimental platform,and can achieve high detection accuracy.