An estimation of coal density distributions by weight based on image analysis and MIV-SVM

Density distribution is an important coal quality index used in the coal industry. The traditional method is excessively complex and time-consuming. Therefore, a new and fast method for coal density distribution estimation by weight is proposed. A semi-automatic local-segmentation algorithm and a multi-scale image segmentation algorithm based on a Hessian matrix were used to identify coal particles. Fifty color and texture features of particle surface were extracted. Mean Impact Value (MIV) and Support Vector Machine (SVM) were applied for feature selection and density prediction. Finally thirty two features were used to establish the estimation model of coal density fractions, and Coal density distributions by weight were estimated by mass predicting model. Ten tests were carried out to verify the application accuracy. Comparing the mean density distribution to the actual density distribution, the absolute errors are 8.66%, -6.33%, -4.06% -2.03% -3.96%, -1.77%, and 9.50% for seven density fractions.