Examinations on the Automatic Classification of Lamellar Graphite Using the Support Vector Machine

The different structure of graphite in cast iron is significant for the mechanical properties of this material. This is why six general forms of a graphite structure, amongst them five configuration classes for lamellar graphite, were defined in the standard EN ISO 945:1994. The subjective classification of the latter can give rise to contradictory results and should be substituted for by an objective classification method using image analyses. The use of the support vector machine is introduced by which a binary classification is performed by calculating a maximum-margin separating hyperplane in the m-dimensional space. The location of the hyperplane is defined by support vectors which are determined by a measurement of image analytical parameters on training images. Six stereological parameters and 14 Haralick parameters were calculated for each image during this examination. The training was based on 350 training images classified by experts. The performance measure of interest was the classification quality of the support vector machine. Furthermore, it was possible to determine the relevance of the 20 parameters for the classification of the individual graphite structures. The support vector machine appears as an interesting classification method from the field of Machien Learning, which can also be employed more generally for metallographic images in the future.