Classification of inertinite in coal macerical based on virtual sample

With respect to the fact that samples of inertinite macerals of coal is not rich enough that makes the classification difficulty, a Gaussian distribution based method is employed to generate virtual samples to expand sample set. Firstly, some initial samples are selected randomly to construct a Gaussian distribution. Then, according to the constructed Gaussian distribution, virtual samples around the initial ones are randomly generated and added to the initial training set. With the expanded training set, a support vector machine (SVM) is lastly built. Results of maceral classification show that: after adding virtual samples the classification accuracy is obviously improved; when the number of virtual samples is add to a certain extent, the classification accuracy tend to stable; virtual samples for small initial training set are more effective than those with big one.