Research On Real-time Detection Method Of Highlighted Complex Surface Defects Based On Machine Vision

High-brightness surface refers to the smooth surface of parts with extremely high reflectivity,almost mirror reflection,which can clearly reflect the image of objects,such as the polished surface of metal parts,the plating surface of plastic parts,etc.With the rapid development of economy,people's requirements for the appearance of products are also increasing,resulting in more and more products with high-gloss surfaces.However,during the production process of such parts,the surface is prone to defects such as scratches,pits,foreign objects,air bubbles,etc.,which leads to a decline in the appearance quality of the product.For some high-precision components,the presence of surface defects on the product will not only damage the performance and life of the parts,but also cause serious accidents and cause significant economic losses.Therefore,accurate detection of surface defects is crucial.At present,the research on the visual inspection of the surface of the high-gloss surface products is mainly concentrated on the high-gloss surface products with regular shapes such as flat or spherical surfaces.It is also the difficulty in industrial testing.In the imaging of traditional visual inspection,due to the highlight characteristics of the part surface,it is easy to produce problems such as local overexposure or surrounding environment mapping,which brings greater difficulty to subsequent image analysis.In this regard,this paper proposes a real-time online detection method for surface defects of bright and complex curved surfaces based on machine vision.Using laser as the light source,using the highlight characteristics of the surface of the test piece,the laser image reflected on the screen through the surface of the test piece is analyzed as the test image,so as to judge whether the surface of the test piece is qualified,so as to establish a new For the rapid detection method of highlighting the surface of complex curved parts.In this paper,based on the characteristics of high-brightness surface,complex appearance and large surface area,the hand shower in bathroom products is selected as the research object.The specific research work includes:First,determine the test standard and overall plan for the test piece.Starting from the key technical problems and difficulties of highlighting complex curved surface detection,multiple solutions are proposed,simple experiments are performed andcompared to determine the optimal solution,and then the overall detection scheme of the detection system is determined.Secondly,after determining the image acquisition scheme,build a detection platform to lay the foundation for the next experiment.From the perspective of imaging effects,this article considers the influence of the relative positions of the light source,the measured object,the camera,and the screen that receives the reflected laser light on the imaging effect to determine the overall layout of the detection system hardware.Then according to the actual needs,consider the main parameters of each component,choose the standard part of the appropriate model,and design and customize the non-standard parts.After the hardware system of the experimental testing platform is set up,the control system is designed and debugged to ensure that the testing platform can meet the requirements of the experimental use.Again,design the algorithm.After image collection through the detection platform,the images are classified and marked according to the requirements of the detection algorithm.In this experiment,the images are divided into two categories,namely qualified and unqualified.Using the marked images as input data sets,complete the learning and training of deep learning algorithms through HALCON,and obtain a classification network that can meet the actual detection needs.In this paper,simulation experiments are carried out according to the actual detection situation.A total of 1578 sample images are learned and trained.After the trained network has tested 237 sample images,the classification accuracy rate can reach more than 96%.Judge that the speed of an image is about 200 ms,which can meet the detection needs in the production process.Finally,summarize the research work of this article and put forward further research suggestions.