Research On Single Crystal Silicon Diameter Measurement Technology Based On Image Processing

Silicon single crystal is the most important basic functional material is the mainstay of the microelectronic and photovoltaic industries.At present,the domestic production technology of silicon single crystal can not reach the foreign level.Most of the domestic production equipment of single crystal silicon is imported from overseas,which is expensive,difficult to maintain and has some technical reservation.Therefore,the domestic production technology of silicon single crystal needs further innovation and research to meet the increasingly stringent industrial requirements.The main innovations and research contents of this paper are as follows:(1)In view of the requirements for the automation of single crystal silicon diameter measurement technology and the current situation that the single crystal silicon image cannot automatically extract the region of interest,This paper focuses on the automatic extraction of single crystal silicon ROI,studies the gray distribution law of single crystal silicon image,creates an objective function S,sets the threshold T of the slope of the objective function,and records the gray level T0 when the slope of the objective function S is less than T for the first time,with T0+(max?gray-T0)/2as the automatically selected binarization threshold.Then label the connected regions of the entire image using the bwlabel function.The smallest rectangular module of the region of interest is determined by obtaining the largest connected area,and the imcrop function is used to extract the ROI of each single crystal silicon image,and finally realize the automatic extraction of the ROI of the single crystal silicon image.This paper processes 300 images of silicon single crystal,of which 282 images can completely be extracted the target area,that is,the accuracy of the ROI automatic extraction algorithm proposed in this paper can reach 94%.In the correctly extracted ROI image,the ROI area accounts for 8.2% to 14.2% of the total image,which can remove 85.8%-91.8% of the invalid background area.(2)In the edge detection stage,in view of the effect of high temperature on thenoise of the image,a combination of image binarization and morphological operation is proposed to eliminate it.Aiming at the “coarse” edge that is common in pixel-level edge detection,this paper proposes Harris corner detection that can detect the grayscale transformation of pixels in various directions.Finally,it can remove 90% of the scattered noise,halo noise and redundant edge points in the original image detection edge points.In order to improve the accuracy of edge detection,the traditional Zernike moment algorithm is improved on the inappropriate discrete template and low sub-pixel threshold efficiency,and finally an edge detection accuracy of less than 1 pixels is obtained.(3)When fitting the ellipse,considering that the upper edge of the energy release ring is noisy and cannot fit the real edge,the upper edge removal algorithm is designed for the crescent-shaped edge point I obtained by the edge detection,and the lower edge point set I1 is obtained.The accuracy of the upper edge removal algorithm can reach 91.3%.Then choose the simple and fast least square method for ellipse fitting,and the accuracy of calculating this method for ellipse detection reaches 88%.In view of the large deviation of the fitting results,the reason for the deviation is analyzed,and the ellipse semi-minor axis length b is proposed as the threshold.When the distance s from the edge point to the center of the fitted ellipse is less than the threshold,that is,when s<b,the edge point is eliminated.Finally,two different single crystal silicon image sets R1 and R2 are tested,and the elimination algorithm is obtained to improve the accuracy of ellipse fitting to 93.3%.And the average floating range of the detected diameter value is within 6.95 pixels.