Coal density distribution is one of the most important production indexes in coal preparation processes, but its traditional measure is time consuming. A fast and efficient method to estimate the density fraction of coal particles becomes an urgent problem. This article proposed a prediction method of density fractions of coarse coal by use of kernel methods and machine vision. Coal particle images were segmented and identified by a multi-scale image segmentation algorithm based on the Hessian Matrix. Thirty-two features, including gray information and texture information, were extracted. The 3σ principle in statistics was applied to remove the abnormal points existing in features of each density fraction, and then all features were normalized. Kernel principal component analysis was used to reduce the data dimensions and the first two principle components were determined as the input of the support vector machine to predict the density fractions of coal particles. The best support vector machine parameters of c and g were determined by the method of K-fold cross validation. Through five tests, the rediction accuracy of training data reaches 84.67%, and the prediction accuracy of test data reaches 81.2%. Results indicated that the kernel method of kernel principal component analysis and support
vector machine is able to predict the density fractions of overlap coarse coal.