Research On SEM Image Segmentation Of Atmospheric Particulate Matter

Atmospheric particulate matter has become one of the most principal pollutants affecting air quality in China.Because different shapes of atmospheric particulate matter have different kinds of harmfulness to the people's healthy,it has been valuable to study the morphological characteristics of particulate matter.Segmentation of particulate matter image is an important prerequisite for studying its morphological characteristics.In this thesis,segmentation of SEM(scanning electron microscope)images of particulate matter was studied.The main research contents are as follows:(?)Two particulate matter samplers,named as PQ200 and 2050,were used to collect samples.The SEM images of particulate matter were obtained by high resolution scanning electron microscope.And the SEM images were preprocessed to reduce noise and to improve contrast.(?)In order to segment images of spherical particulate matter with cover,this thesis optimized the sampling strategy and proposed an iterative diameter circle detection algorithm.The boundaries of particulate matter images are obtained by iterative diameter sampling strategy.Refinement method is adopted to obtain possible circles.Possible circles are detected to obtain candidate circles.Finally,the images of spherical particulate matter are segmented using candidate circles.The iterative diameter circle detection method not only effectively reduces the number of candidate circles,but also improves the detection accuracy of the center and the radius of the circle,which secured the segmentation of the images of spherical particulate matter without cover.(?)In order to solve the problem relating to segmentation of spherical particulate matter images with cover,the edge detection function was improved,and geodesic active contour model was combined with gray co-occurrence matrix.The initial contour is evolved using the level set method to obtain the target boundary.The difference between the angular second moment in the neighborhood of the boundary pixel and that in the target area is calculated.If the difference is greater th an the threshold value,the point would not be the target boundary,and the evolution curve would move to the center of the contour.If the difference is less than the threshold,the point would be considered as the target boundary.Compared with geodesic active contour model and active contour model based on deviation compensation,the improved geodesic method optimizes the evolution process of contour curve and realizes the segmentation of spherical particles with cover.The experimental results show that the proposed method not only effectively improves the accuracy of contour detection,but also achieves segmentation of particulate matter images,which lays a foundation for the subsequent study of particulate matter morphological characteristics.