A Method For Quickly And Accurately Recognizing EBSD Pattern

EBSD(Electron Backscattered Diffraction),a necessary in scanning electron microscope,has been widely used in Materials Science,Earth Science and other fields for which combines the advantages of the X-ray analysis and the electron diffraction in transmission electron microscope.Based on such a distinguishable feature,EBSD technique plays an extremely vital role in characterizing the micro structure of crystals.The material area to be characterized consists of 10~7-10~8 pixels in most cases,and a single EBSD pattern can be extracted from each pixel.According to the EBSD patterns,the orientation of the crystal can be determined by taking consideration of the Kikuchi bands recognized from patterns with the aid of Hough Transform(HT).It is clear that the result of the characterization can be affected by the quality of EBSD pattern while the efficiency of that is subject to HT.The EBSD patterns usually have low brightness and uneven gray level,this paper developed a program for background deduction to enhance the brightness of the patterns in order to sharpen the contrast of the Kikuchi bands extracted from the patterns.Surprisingly,the result shows that the program not only can enhance the brightness of the patterns,but also has the ability to improve the gray level.Compared with commercial software from Oxford,the Kikuchi bands with much more sharpen contrast are available according to this program and the image quality of the EBSD patterns is much closer to 0.5.To verify the influence of the background deduction algorithm on the recognition of Kikuchi bands,this paper respectively employed HT in the original EBSD pattern,the EBSD pattern under background deduction and the EBSD pattern processed by Oxford company.After comparing the recognition results of the three in detail,it is found that the Kikuchi bands of the original EBSD pattern can be hardly recognized,but the recognition accuracy of Kikuchi bands in EBSD patterns has been significantly improved both whose background is deducted and which is treated by Oxford company.Through background deduction,not only pattern brightness can be enhanced,but also the accuracy of Kikuchi bands recognition can be improved.Thus,it is an extremely important step to deduct background in EBSD patterns processing.Moreover,the effect of the number of pattern pixels on the HT time is explored in this paper.It turns out that the more the number increases,the longer the time grows.The time of HT will increase linearly with the number increase of pixels in the target pattern and even at an exponential level with the pixel increase in the original pattern.This paper is based on Central Processing Unit(CPU)to perform HT and discusses the influence of CPU thread number on HT time.The results show that the time will drop with the CPU thread number rises.When the number of threads is less than of cores,the thread acceleration ratio increases linearly with the increasing thread number.When greater than the number of cores,however,the upward trend of thread acceleration ratio slows down.By utilizing Compute Unified Device Architecture(CUDA),the HT time can be accelerated.And the influence of pattern pixel number on HT is also analyzed in this case.The HT time increases with the increasing pixel number in the original pattern increases.In the meantime,both the data transmission and calculation time show growth.The transmission time rises greatly but the acceleration ratio of CUDA decreased;As the number of pixels in the target pattern increases,the HT time increases during which the data transmission time of CUDA remains unchanged,the calculation time increases,and the acceleration ratio of CUDA increases.At the same time,several graphics cards are used to perform HT.The HT time will increase but the calculation proportion of CUDA will decrease when the number of cards and CUDA computing platforms grows.Finally,with the single CPU and the single CUDA,the HT speed and Kikuchi bands recognition are compared respectively.It is shown that the recognition results of Kikuchi bands is the same,but the HT speed with CUDA is faster.The diffraction geometry of EBSD determines that the traces of all the Kikuchi bands passing through the same zone axis should meet precisely at one point.The recognized traces belonging to a zone axis,however,do not exactly intersect at the same position since the Kikuchi bands are independently detected one by one without any restrictions in the HT.Therefore,this paper utilizes the rotational symmetry to localized the position of zone axis from an EBSD pattern and match the traces in order to correct the determined result of the HT.The pattern center(PC)and the detector distance(DD)have small influence on the accuracy of the position of zone axis.Even if there exists error between the PC and the DD,the position of the zone axis can still be accurately localized by utilizing the rotational symmetry.In the meanwhile,the PC and the DD can be approximately estimated according to the maximum value of the similarity of zone axis,thereby correcting the data error in the experiment.In addition,this paper uses the simulated EBSD pattern and the experimental EBSD pattern respectively to verify the traces.By calculating the deviation of the zone axis and the traces,the results show that the corrected traces can reduce the deviation of the zone axis and the trace and thus accurately recognize the Kikuchi bands in the EBSD pattern.This paper discussed a new method for quickly and accurately recognizing EBSD pattern on the basis of HT,which is valuable for the development of the EBSD technique and the domestication of the scanning electron microscope.