| ZnO nanowires are widely used in various fields because of their excellent physical and chemical characteristics.The size,shape,and distribution of ZnO nanowires are important parameters affecting their physical and chemical characteristics.How to accurately measure the characteristic parameters of ZnO nanowires according to SEM image,are of great significance to the relationship between the microstructure characteristics and the physicochemical characteristics of nanomaterials,and in particular,they can promote the theoretical understanding of their physical mechanisms.However,the currently used measurement methods are manual measurement.This method is low efficiency,high cost,and cannot be automated.Since deep learning was proposed in 2006,image processing methods based on deep learning have developed rapidly,and their performance in many image recognition fields is far superior to traditional image processing algorithms.According to the characteristics of ZnO nanowire images,this paper proposes an image processing algorithm based on deep learning,which can effectively and accurately segment images.The main contents of this paper include the following two parts:(1)Construction of secondary electron image datasets for ZnO nanowires.Currently,ZnO nanowires lack effective public image datasets for microscopic images,and it is difficult to meet the basic requirements of deep learning algorithms for large amounts of image data.To this end,it is proposed to use automated and batch growth equipment(codenamed KQS system in the laboratory)to prepare a large number of ZnO nanowire samples in different states,and improve the validity level of regulating the structural parameter characteristics of samples in the same state.Compared with the traditional method using manual experiments,the KQS selfevolving process machine can achieve 32 samples in one experiment,which greatly reduces the time cost and greatly improves the experimental efficiency.In addition,the level of data traceability is improved,and Reduce the impact of human error.(2)Example segmentation task for the secondary electron microscopic image of ZnO nanowires.The segmentation task of nanowire images mainly has the following three difficulties: 1)The nanowire SEM image is a singlechannel grayscale image,and the gradient distribution of the image is not obvious.Traditional methods such as pixel thresholds and clustering operations are used to segment the image,and the effect is not obvious;2)the SEM images of nanomaterials have different shapes,and the same material,based on different parameters,will have completely different characterizing images,which requires the algorithm It is very robust and can process SEM images of ZnO nanowires in different scenarios;3)The nanowire SEM image is a collection of some nanowire monomers.The way in which nanowires are stacked and arranged has high complexity and randomness,and the monomers overlap each other,which causes strong interference to the feature segmentation task.This article analyzes and discusses the algorithm principle in view of the above difficult problems,and conducts a lot of experimental research work.Based on this,a Mask RCNN-based image segmentation algorithm was proposed.In the 48 samples provided by the KQS system and 146 SEM images of ZnO nanowires with various shapes,a recall rate of 86% was successfully achieved,and about88% Feature effective area,reducing false recognition rate to 14%.The measurement error of the end surface area and perimeter of the ZnO nanowires is 3.6% and 6.3% respectively. |
PDF Full Text Request