Study On 3D Reconstruction Of Weakly Textured Micro-Objects Under Scanning Electron Microscope

Due to the advantages of high resolution,scanning range and large depth of field,Scanning Electronic Microscope(SEM)is widely used in the measurement and evaluation of various micro-nano targets(such as micro-electromechanical systems and integrated circuits).However,due to its imaging principles,current SEM-based observations are often limited to in-plane and cannot directly obtain depth information.Driven by application requirements,the visual 3D reconstruction technology based on optical sensing at the conventional scale has developed rapidly in recent years.However,limitated to optical resolution and small depth of field in optical microscope,3D reconstruction techniques in conventional scale optical directly apply to micro-nano targets become more difficult.To this end,the article will study the threedimensional reconstruction under SEM,which used as the imaging equipment,MEMS devices and microchips as main object,which with no obvious texture features on the surface and the feature size distribution in the micron level.The main contents of this article include the following:(1)Take the multi-scale calibration board as the object,develope a calibration plate feature point extraction algorithm based on morphological processing,realize the calibration of the SEM visual imaging model.First,the imaging principle of the SEM and the visual imaging model at low magnification are introduced;then,for the difficulty of extracting the corner points of the multi-scale calibration plate,feature point extraction algorithm is developed;finally,the classic Zhang Based on the standard calibration method,the calibration of SEM is completed and verified by experiments.(2)Taking the five-axis precision positioning stage in the SEM as the object,first establish the kinematics model of the five-axis precision positioning stage,and realize the calibration of the geometric parameters of the kinematics model through nonlinear optimization.Then,the coarse positioning of the platform is realized based on the kinematics model after calibration,and the precise positioning of the platform is realized based on the autofocus curve,so as to realize the automatic acquisition of the multi-view image sequence.Finally,the sensitivity analysis of the factors that affect the positioning and model calibration results is carried out through simulation,and the positioning accuracy of the kinematics model after calibration is within 5 microns,which meets the accuracy requirements of image sequence acquisition.(3)Aiming at the characteristics that the surface texture characteristics of weakly textured micro-targets are not obvious,this paper proposes an improved ORB(Oriented FAST and Rotated BRIEF)feature point matching algorithm to solve the problem of high mismatch rate caused by similar textures.The improved algorithm is based on ORB for feature point detection,LK(Lucas-Kanade)optical flow tracking algorithm for feature point matching position prediction,and then based on KNN(K-Nearest Neighbor)algorithm for matching candidate points,effectively improving feature points Matching efficiency.Finally,the classical feature point matching algorithm is compared with the improved feature point matching algorithm proposed in this paper.Experimental results show that the performance of the improved feature point matching algorithm proposed in this paper has been effectively improved.(4)Based on the previous visual calibration results,image sequence acquisition and improved image matching algorithm,the SEM image sequence of weakly textured microobjects is reconstructed in three dimensions through the incremental motion recovery structure(Structure From Motion,SFM).First,the principle and system of incremental 3D reconstruction are introduced.Then,the improved matching algorithm and the classic matching algorithm proposed in this paper are used in three-dimensional reconstruction experiments,respectively,to verify the effectiveness of the proposed feature matching algorithm.Finally,by comparing with two commonly used 3D reconstruction software,the effectiveness of the proposed 3D reconstruction method for 3D reconstruction of weakly textured micro-objects under SEM is further verified.