A Recognition Method of Mineral Shape Based on Extreme Learning Machine

Abstract

In view of the situation of the existing algorithm for mineral shape recognition is relatively complex, the individual of strong pertinence and poor robustness, the use of infrared thermal images of minerals multifractal feature data classification recognition method is put forward. Multifractal can describe not only the local details, but also the overall characteristics that has the scale independence and theoretically is suitable for describing the texture characteristics and the distribution of mineral as well as that of energy resource. This paper uses multifractal as parameters of singularity detection of high-dimensional data and learning and understanding of high-dimensional data to distinguish the object/target from infrared heat map. The experimental result show that the infrared thermal image of mineral target in line with the multifractal characteristics, which can be used as one of the effective methods of infrared thermal images detection target. When three kinds of neural network ELM, PNN, GRNN is used for machine learning with obtain fractal parameters, ELM’s accuracy is as high as 84%. While the same training with face natural images is done, ELM is still best, but accuracy is less than 15%. It shows that ELM combining with mineral fractal data has a better performance in classification and pattern recognition.