Research On Grade Estimation Method Of Iron Ore Based On Image Features

Real-time iron ore grade estimation has been the key technical problem of mining operation sectors for a long period of time. As the traditional detection methods of mining enterprises, after many years of explorations and improvement, chemical analysis has many advantages, such as mature technology, standardize operational processes, accurate test results, etc; but it also has some disadvantages, such as slow information feedback, high cost of detection, poor real-time performance, etc. The paper tries to put forward a real-time method for iron ore grade estimation. It has important significance to increase the production efficiency of mining enterprises.Taking eighteen kinds of iron ore samples as the research object, the feasibility of applying the image features in iron ore grade estimation has been corroborated by a preliminary experiment. To be clear, the grade of these iron ore samples was established by chemical analysis from Nanjing Meishan Mining Company. Concrete research content including following several aspects: the first is overlapping segmentation of iron ore images; the second is feature extraction and selection of iron ore images; the last is establishment of valuation model for iron ore grade and the analytic results of iron ore grade estimation.Aimed at the characteristics of the iron ore mutual adhesion in the image, the paper presents an algorithm for separation between ore targets, which is based on local average with double thresholds. Several techniques, such as hole-filling, distance transform, and morphological reconstruction were used to obtain the ore seed region.Watershed transformation was then applied to effectively segment the ore object.Iron ore images of different grades were directly used as samples. The statistical feature based on gray histogram and the textural feature based on the gray level co-occurrence matrix, run-length matrix and wavelet transform were extracted from each iron ore image to obtain a sample set Sample1. The subregions after iron ore images segmentation of different grades were used as samples. The statistical feature based on gray histogram and the textural features based on the gray level co-occurrence matrix were extracted from each subregion to acquire sample set Sample2.PCA and Isomap algorithms were used respectively on two groups of sample set for feature selection. According to the accuracy of the final ore grade estimated results to determine what kind of feature selection methods.Finally, in accordance with the two groups of sample set Sample1 and Sample2,the valuation model for iron ore grade was established based on SVM and RBF neural network,respectively. The estimated effect of two iron ore grade estimation methodswere compared and analyzed. The research results of this paper will provide an experimental basis for further study on real-time iron ore grade estimation.