The Nano-particle Image Recognition Based On TEM

Nano-technology is a new technique which is gradually developed in the late of 1980s, it is also one of important contents of technological industry revolution in 21 centuries. Nano-composite’s particle size distribution is one of key factors in the impact of material properties, it is one of important indexes about characterizing zero dimentional nanometer materials. How to detect and evaluate the particle size and performance of nano-material has become the most important issue to be addressed in the aspects of zero dimentional nanometer materials’ preparation and application. Transmission electron microscopy (TEM) is an important tool of nano-material’s characterization,under which you can see the submicroscopic structure and ultrastructure less than 0.2um that can not be seen under an optical microscope. Currently, the TEM resolution is up to 0.1～0.2nm. It can be used to observe the fine structure of the sample, even only the atomic structure.In this paper, we use the digital image processing knowledge to do a more comprehensive understanding and analysis obout the theory and algorithms of particle recognition technology in recent years, and focuses on doing a depth study and research on three key algorithms of nano-particles recognition technology include background segmentation, particle adhesion segmentation and parameter statistics. On this basis, using the Java programming language, we design a practical software that have a set of functions include threshold segmentation, overlapping segmentation and achieve automatic statistics of nano-particle’s parameter such as the size, the number, roundness, etc.The main works of this paper are described as following:First of all, we make a simple introduction and analysis about TEM in order to understand the characteristics of the TEM images better.Through the image preprocessing technology such as smooth work carries on the preliminary processing of image combining the characteristics of a TEM image,improve the quality of the TEM image, and lay a foundation for the subsequent processing and analysis of nano-particles images.For particles recognition of the TEM nano-particle image,the segmentation of the particles and the background is the basis and difficulties of the entire identification process.In this paper, the maximum correlation criterion method is used to the segmentation of particles and background by comparing the advantages and disadvantages of a variety of background segmentation method and combining the characteristics of TEM nano-particles images. In the segmentation process when the target pixel is close to the background pixel, it’s easy to divide the target into the background, and cause internal particles appearing holes that should to be filled.Due to the particularity of TEM nano-particles image, there is always a kind of phenomenon that a small number of particles stick and even reunite together. It’s a difficulty of nanometer particles recognition system for segmenting touching objects. This paper studies the traditional watershed algorithm and common watershed transformation algorithm, by comparing and analyzing their advantages and disadvantages in the segmentation of linking particles, we use a method which is a combination of fast watershed transform and region merging to achieve segmentation of the linking particles of the image finally. The segmentation result is better to segmentalize the linking particles accurately. The processed image can achieve the statistics about the number, the size (area), and perimeter of particles better.The last is the statistics of nanoparticles’ parameters. This article uses the recursive method for calculating the area and the number of particles, chain coding is used to calculate the particle’s perimeter.Software system implementation is to use the Java programming language, on the basis of cross-platform design and realize the recognition of TEM nano-particles’ number, size (area) and roundness, etc. such a set of parameters’statistical work. By testing the system show that:the system can complete the import of the TEM image of nano-particles and achieve the recognazition of TEM nano-particles and statistical work of parameters, it works well in our experiments. The system runs stably, through comparing with the artificial statistics, the whole automatic recognition rate is 88% when testing the system.The nanoparticles image recognition system designed in this article is easy to use, it has faster recognition and higher accuracy, and it has a good application prospect.