Study On Grain Detection Method Based On Machine Vision

With the development of computer technology,artificial intelligence technology,image processing and recognition technology,machine vision technology plays an increasingly important role in various fields of national economy,especially in modern intelligent manufacturing industry.Machine vision technology brings technical innovation to traditional industrial automation field.And it has greatly improved the quality and productivity of automation products.In the process of semiconductor material processing,the quality of grain directly determines the quality of wafers and electronic components.Besides the qualification test of electrical parameters,the detection of physical defects is also very important.At present,most enterprises use manual detection of physical defects,but there are some following disadvantages:firstly,the labor cost is high,and it is easy to get visual fatigue when doing the boring work with strong repetition for a long time,which leads to the production efficiency reduction.Secondly,the wafer quality detection can also be affected by the subjective factors of the skilled workers,resulting in incorrect judgment and lower yield.In addition,due to the wafer is very thin and brittle,if contact detection is adopted,the damage rate will increase,the accuracy and speed of detection are difficult to be guaranteed.Therefore,in order to solve the above problems,a method for detecting grain physical defects based on machine vision is proposed,which mainly focuses on the research of wafer correction,target positioning,grid reconstruction and traversal detection.In this paper,the detection algorithm is studied by OpenCV Visual library in Microsoft Visual Studio integrated development environment,and implement the wafer image correction and positioning by C++programming language design.We have also proposed the method of adaptive wafer grid reconstruction.Comparing the mean value of grain binary figure with a set threshold value,the physical damage of grain surface can be determined.After image binarization processing,it's convenient to determine whether there is a stain on the grain surface by judging the number of pixels.For the oxidized grain,we know whether it was oxidized by comparing its average gray value with a threshold which is determined by experiments.After the grains on one wafer are identified over,the parameters such as the coordinates,area and identity mark are recorded to a txt file.Using SQL Server database to create the wafer database,and the GUI of database is designed with VB.NET programming language.The feasibility of the study method is verified by image detection of the actual wafer,and the defect detection effect is good through visual display of the detection process.This subject presents a new calculation method of wafer correction angle is proposed.Firstly,the linear detection of the wafer image is carried out by Hough transform based on the cumulative probability.Then,the angles between the straight lines and the X-axis are clustered,and the average value of the clustering results is taken as the correcting angle.At the same time,an adaptive wafer grid reconstruction method is proposed.For grains of different sizes,the corresponding grids can be automatically reconstructed by clustering the outer rectangular horizontal and vertical directions of grain contour.In order to overcome the shortcoming of the artificial detection of grain defects,this paper puts forward the detection scheme of machine vision.Based on above research work,the technical challenges and improvement suggestions are pointed out.