Abstract: With the wide application of digital recognition technology in image analysis under the microscope,the intelligent recognition of substance type under the microscope has become a basic problem of microscopic analysis. Aiming at the problem of low precision of mineral intelligent recognition in image,a multi matrix fusion machine learning intelligent recognition model was constructed by taking color matrix,texture matrix and RSTC moment invariant as recognition characteristics and entropy weight and coefficient of variation weight as initial recognition weights. In this paper,the image sets of magnetite,mica, calcite,brass and calcium ferrite were selected as test samples,and the characteristics of color matrix,texture matrix and RSTC moment invariant were extracted. The contribution rate of features in image recognition was quantitatively analyzed,and the intelligent recognition experiment of multi-matrix fusion machine learning was carried out. Test results show that the contribution rates of different types of feature indexes in the process of image recognition are significantly different,the machine learning intelligent recognition model based on multi matrix fusion has good recognition rate and robustness,and can significantly improve image recognition accuracy. Index entropy weight and variation coefficient class weight as initial weight can obviously promote the rapid convergence of the algorithm and reduce the recognition time.

Key words: mineral image,multi-moment fusion feature,intelligent identification,comprehensive weighting