Research On Surface Defects Visual Inspection System And Algorithm Of Polished Tile

China produces and exports a large number of building tiles every year.Society’s demand for ceramic tiles is growing rapidly,which causes that the requirements for the production quality of ceramic tiles are getting higher.Among them,polished tiles have become the mainstream products in the tiles production industry because of their bright color,resistance to abrasion and corrosion and many other advantages.The detection of surface quality is a necessary process in the manufacturing process of polished tiles.At present,domestic ceramic tiles production producers have achieved localization and automation of all kinds of equipment in ceramic tiles production line except for quality detection,but most producers still rely on traditional manual detection method to detect the defects,which is not only inefficient,but also difficult to meet the consistency requirements of detection,and can not adapt to the development trend of high-speed intelligent ceramic tiles production line.Relying on the powerful ability to feature extraction and excellent generalization performance,deep learning technology achieves great success in many industrial inspection tasks,but there are few related researches applying it to the defect detection of ceramic tiles’ surface.Based on the existing surface defect detection technology and deep learning technology,this paper conducts in-depth research on the related technical details for the surface defect detection of textured polished tiles,then proposes a complete surface defect detection plan for polished tiles based on deep learning.The main contents of this paper are as follows:(1)Aiming at the problem that the surface defects of some polished tiles are subtle and difficult to highlight in the imaging process,first the surface optical characteristics of polished tiles are analyzed,and then combined with the actual production environment as well as detection requirements,a real-time image acquisition system of polished tiles based on linear array CMOS is designed.By selecting the appropriate hardware and lighting method,which bases on dark-field lighting method of bright linear LED,the system can stably capture the surface image of polished tiles with high quality.(2)Aiming at the problem of uneven gray level along the horizontal resolution direction of image caputured by linear CMOS image acquisition system,the causes of uneven gray level is analyzed,the a gray level uniformity correction method combining pixel response characteristic correction and gray level mean subtraction correction is proposed,which can effectively improve the gray uniformity of the image whose results is evaluated by quantitative indicators.Aiming at the problem that the noise produced in the acquisition process of CMOS image will interfere with the subsequent detection,the causes and manifestations of the noise is analyzed,for which an appropriate spatial filtering denoising method is selected through experiments.(3)Aiming at the problem that the defects in polished tiles are subtle and it is difficult to detect them stably and effectively by traditional image processing methods,first the target detection algorithms is studied and then two-stage Faster-RCNN network and one-stage yolv3 network is compared to analyze their function in defect detection.Finally,according to the detection requirements of the system,yolv3 is selected as the basic frame of the detection algorithm.In order to make the network more accurate in detection,a convolutional autoencoder is inserted into yolov3 to reconstruct the weak-defect image of the input to enhance the input of the network.Results reveal that the proposed network can identify the holes,pinholes and scratches defects on the surface of polished tiles effectively,with a m AP of 83.5%.(4)Based on the analysis of the actual demand of production and the defect detection algorithm based on deep learning,a visual detection software for surface defects of polished tiles is developed,which contains image acquisition module,defect detection module and statistical display module.Some experiments were taken to test the feasibility of the proposed detection system.The actual detection results of the polished tiles surface images show that the proposed algorithm can stably detect the surface defects with width over0.8mm of polished tiles.The detection time for a single 600 mm * 600 mm polished tile is about 7S.