Research On Image Segmentation Algorithm In Van Der Waals Heterostructure Construction System

In materials science,van der Waals heterojunction is a very popular field.The van der Waals heterojunction is formed by stacking two-dimensional materials on demand,and its excellent material properties make it of great research significance.At present,the construction of the Van der Waals heterojunction needs to rely on manual manipulation of mechanical equipment for construction,which is inefficient and low in construction complexity.In response to this problem,the Hefei Research Institute of the Chinese Academy of Sciences set out to develop the Van der Waals heterojunction precise construction system,in which the rapid identification of two-dimensional materials is a key part of the system.In order to meet the needs of the system for rapid identification of two-dimensional materials,combined with the widespread application of artificial intelligence in the field of computer image processing in recent years.Under this background,this paper focuses on the research direction of rapid identification of two-dimensional materials,and conducts research on two-dimensional material segmentation algorithms based on deep learning.main tasks as follows:(1)This article builds an automated scanning platform.The platform can scan the prepared two-dimensional material samples in all directions and collect images of the samples for later identification of the two-dimensional materials.The automated scanning platform can completely collect sample images.At 500 times magnification,a sample size of 1 square centimeter can collect at least 12,500 images,avoiding the huge workload and omissions during manual search.(2)This paper produced a two-dimensional material segmentation data set.A large number of sample images were collected using the built automatic scanning platform.After screening,an effective image containing two-dimensional materials is obtained and the data set is produced.The data set contains 10,000 training set images with segmentation labels,and 100 test set images for testing the performance of the algorithm.(3)This paper implements a deep learning algorithm suitable for two-dimensional material image segmentation.Because the features of two-dimensional material images are complex and difficult to capture,Traditional graphics algorithms are not suitable for two-dimensional material image segmentation.This article uses deep learning algorithms to improve on the basis of SegNet network,in order to extract the characteristics of two-dimensional materials.By applying the underlying features,it compensates for the problem that the SegNet network output feature map loses detailed features,and improves the problem of poor target edge segmentation when SegNet network performs image segmentation.Experimental results show that the improved algorithm is suitable for two-dimensional material segmentation,and its performance has been greatly improved.