- مبلغ: ۸۶,۰۰۰ تومان
- مبلغ: ۹۱,۰۰۰ تومان
Detection of size, shape and color of minerals are important for obtaining information about minerals. The output of mines is ores which vary in colors and shapes. The multiplicity of ores, large scale features and the importance of speeding up the mineral type detection process for intelligent systems, leads us to rely more on expert's advice and rank the selected available features for type detection, according to their importance. In this paper, to separate different ores and gangue minerals, image processing and computer vision techniques with combination of multi criteria decision making (MCDM) approach are applied. Our method proposes a novel way which combines the image processing techniques and artificial neural networks, with analytic hierarchy process (AHP) approaches to detect different types of ores. By help of experts in feature ranking, the image processing techniques proved to be more effective and prompt. The final results show that the proposed method is more successful in type detection of minerals than the other image processing techniques for ores type detection. Our method is also applicable for real-time systems to estimate minerals at on-line ore sorting and classification stages.
The use of image processing in advanced process control systems is an enabling technology in the mining and minerals processing industry, with a wide range of potential applications. In this study, we designed a new system for the categorization of ores and minerals extracted from the mines, using artificial neural networks and analytic hierarchy process. The proposed approach significantly outperformed other methods such as [8,12]. Significant improvements was shown by introducing combinations of AHP ranking and image processing techniques along the ANN structures to enhance the estimation of rock types present in the mixture. The reported performance suggests that this approach could be deployed in on-line ores type detection stations to assist operators in the detection of different types of ores. Due to its generic nature, the proposed method can be used to detect many classes of ores even when only a modest dataset of examples is available. The proposed framework has been extensively evaluated on number of ores images to ensure the accuracy of the obtained results. Our experimental results, conducted with sixteen widely used categories of ores. The corresponding features weights are calculated according to experts advice. It should also be noted that the classification accuracies reported for the sixteen considered type of ores are calculated using weighted features.As a resultthe obtained accuracies are 9.3% higher than the other presented methods, in average. This can prove the importance of expert’s comments for using and emphasizing on more influential features which results as a better classification output. The proposed method could be used for automatic on-line rock classification and sorting which in turn could help in optimizing, for instance, the throughput of mills within a mine. Accuracy estimations were also presented for quantitative assessment of the machine vision system. Although this model trained and tested on specific dataset, but the features and the designed model is applicable for other ores data sets to classify other minerals.