Recognition method of coal and gangue combined with structural similarity index measure and principal component analysis network under multispectral imaging

Abstract

Accurate identification of coal and gangue is an important prerequisite for intelligent separation, which can help the intelligent development of coal mines. Based on multispectral imaging, we propose a coal and gangue identification method combining structural similarity index measure (SSIM) and principal component analysis network (PCANet). It is found that the multi-spectral imaging information at the 19th wavelength (914 nm) is most suitable for constructing the identification model of coal and gangue under the SSIM evaluation index. At the same time, the optimal hyperparameters of PCANet model are further determined, that is, PatchSize = 3 × 3, NumFilters = 6, HistBlockSize = 6 × 2, and BlkOverLapRatio = 0. In addition, compared to convolutional neural network (CNN) model, we found that PCANet is less dependent on the number of samples. The results show that the coal gangue identification scheme based on PCANet combined with MSI is feasible and effective, which is helpful to promote the progress of coal gangue automatic separation technology.