ABSTRACT Automatic identification and classification of objects being the results of image recognition algorithms became more and more popular in many aspects of human activity. On the other hand, manual stereological methods and conventional image analyzer are more often than not difficult and time-consuming tool to obtain the informative data especially for complicated microstructures. To solve these problems, a computer assisted quantitative metallographic analysis was explored. The input data for the proposed analysis was a set of digital 2D images of metal microstructures of the technical aluminum cast Al-Si alloys. Images were obtained by high quality cameras embedded in optical microscopes. The objects of interest were the precipitates of intermetallic phases of various morphological shapes. Traditionally, descriptions of microstructures have been based on measurements of topological relationships between the three- dimensonal space and two-dimensional microsections, such as grain size, the average volume of particles, volume fraction, size of particles in unit volume, etc. We consider these features to be insufficient for the process of classification which permits differentiation. Therefore, the computational methods of pattern recognition have been applied to both the statistical particle shape analysis and topological characterization of dendritic structures. Several examples of designed and implemented algorithms, including the measurements of compactness, scale and rotation invariant moments, fractal dimension, convex hull, lacunarity and many other parameters are presented. The key to this quantitative analysis is the manner of interpretation of aluminum alloys' planar microsections. It provides practical techniques for extracting quantitative information from measurements. It is these features that determine the mechanical properties, and any advanced understanding of microstructure- property relations requires their quantitative description. The presented approach is aimed at designing a system for identification and classification of microstructures occurring in multiphase cast alloys. Image data representing diverse samples was taken into investigation. Within each sample alloys’ features were determined based on a cast modeling process. Due to the fact that the presence of specific microstructures determines mechanical properties of cast alloys, an automated image based classification system may be an invaluable tool for developers of modern casting technology. Keywords: computer vision, pattern recognition, image processing, identification of metal phases, quantitative metallography 1. INTRODUCTION Polyphase metal alloys are still the most common structural materials in the production of many goods. Increasing demands regarding on one side, the quality of the products and on the other saving costs and environmental friendly technology are opening the wide fields for advanced methods of material investigations. The starting points of prospective material modification are always very well-known relationships {C,T} ↔ {UP} (where C- chemical composition, T- technology, UP – utilizable properties) [1-6]. However, the another relationships {M} ↔ {UP} (where: M- microstructure) represents a more close and direct interaction which can be implemented into physical or statistical material models [7]. The term ‘material microstructure’ in material science means a 3D construction composed of the particular elements differing in physical, chemical and morphological properties. Light microscopy investigations relate to the 2D representatives of the microstructure constituents, revealed on the metallographic plane cross sections with special preparation procedures. The known stereology relationships allow direct matching of the 2D quantitative global parameters for some microstructure models to their 3D equivalents [8-10]. However, the general description rules for the local features of the material constituents, important from the point of view of its model behavior have not been until yet established [11-13]. Especially, in the case of concave dendritic particles an anticipated 2D ↔ 3D morphology relationship can be univocal and even contradictory (Fig.1). The quantitative description of the local microstructure features as shapes of particular elements is one of the most important and difficult problems in microscope image analysis. A Image Recognition, Identification and Classification Algorithms in Cast Alloys Microstructure Analysis Anna Romanowska-Pawliczek e-mail: [email protected]Department of Applied Computer Science and Modelling, Faculty of Metal Engineering and Industrial Computer Science, AGH University of Science and Technology, Kraków, Poland Aleksander Siwek Department of Applied Computer Science and Modelling, Faculty of Metal Engineering and Industrial Computer Science, AGH University of Science and Technology, Kraków, Poland Miroslaw Glowacki Department of Applied Computer Science and Modelling, Faculty of Metal Engineering and Industrial Computer Science, AGH University of Science and Technology, Kraków, Poland Malgorzata Warmuzek Foundry Research Institute, Kraków, Poland
6
Embed
Image Recognition, Identification and Classification ... Automatic identification and classification of objects being the results of image recognition algorithms became more and more
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
ABSTRACT
Automatic identification and classification of objects being the
results of image recognition algorithms became more and more
popular in many aspects of human activity. On the other hand,
manual stereological methods and conventional image analyzer
are more often than not difficult and time-consuming tool to
obtain the informative data especially for complicated
microstructures. To solve these problems, a computer assisted
quantitative metallographic analysis was explored. The input
data for the proposed analysis was a set of digital 2D images of
metal microstructures of the technical aluminum cast Al-Si
alloys. Images were obtained by high quality cameras
embedded in optical microscopes. The objects of interest were
the precipitates of intermetallic phases of various morphological
shapes.
Traditionally, descriptions of microstructures have been based
on measurements of topological relationships between the three-
dimensonal space and two-dimensional microsections, such as
grain size, the average volume of particles, volume fraction,
size of particles in unit volume, etc. We consider these features
to be insufficient for the process of classification which permits
differentiation. Therefore, the computational methods of pattern
recognition have been applied to both the statistical particle
shape analysis and topological characterization of dendritic
structures. Several examples of designed and implemented
algorithms, including the measurements of compactness, scale
and rotation invariant moments, fractal dimension, convex hull,
lacunarity and many other parameters are presented. The key to
this quantitative analysis is the manner of interpretation of
aluminum alloys' planar microsections. It provides practical
techniques for extracting quantitative information from
measurements. It is these features that determine the mechanical
properties, and any advanced understanding of microstructure-
property relations requires their quantitative description.
The presented approach is aimed at designing a system for
identification and classification of microstructures occurring in
multiphase cast alloys. Image data representing diverse samples
was taken into investigation. Within each sample alloys’
features were determined based on a cast modeling process.
Due to the fact that the presence of specific microstructures
determines mechanical properties of cast alloys, an automated
image based classification system may be an invaluable tool for