Research On Recognition Of Fine-grained Minerals Based On Deep Learning

As an important basic energy in China,coal plays a leading role in the production and consumption of primary energy for a long time.In the process of coal generation and mining,various mineral impurities are inevitably mixed in coal,and environmental pollution caused by the coal combustion process is mostly related to this.Therefore,it is an inevitable choice for the harmonious development of the coal industry and the environment to use the separation and processing method to remove minerals from coal and clean the coal.The particle size distribution of pulverized coal and the mineral component information in coal reflect the overall quality of coal preparation.With the rapid development of artificial intelligence,applying deep learning technology to the process of coal preparation can obtain accurate pulverized coal particle size and its component information,so as to achieve the goals of improving coal preparation production efficiency,reducing production energy consumption and improving resource utilization.The premise of pulverized coal particle size detection is particle segmentation,but traditional image segmentation algorithms are difficult to achieve ideal segmentation results due to problems such as mutual adhesion among particles,uneven particle size distribution and the edge effect caused by scanning electron microscopy acquisition of fine-grained pulverized coal images.In view of the shortcomings of the existing algorithms,this paper proposed a pulverized coal particle segmentation algorithm based on deep learning to accurately segment pulverized coal particle images.At the same time,in order to evaluate the effect of the coal preparation and determine whether the coal preparation has effectively removed minerals in coal,this paper proposed a mineral components recognition algorithm based on deep learning to identify and detect various mineral components in coal preparation products.The specific research work and innovations of this paper are as follows:1.Aiming at the problems of particle adhesion,small particle leakage segmentation and inaccurate edge positioning in the process of pulverized coal particle segmentation,this paper used the Mask R-CNN algorithm based on deep learning to segment the adhesion particles.By using atrous convolution to improve Mask R-CNN to strengthen the multi-scale feature learning of particles,the algorithm effectively solved the problem of small particle leakage segmentation.By improving the loss function in the Mask R-CNN mask segmentation sub-network,the network was promoted to focus on learning edge pixel features,and edge positioning accuracy was improved during particle segmentation.Taking the secondary electron image of pulverized coal particles collected by scanning electron microscopy as experimental data,the experimental analysis of the proposed algorithm showed that the improved Mask R-CNN segmentation algorithm greatly improved the accuracy of pulverized coal particle segmentation.2.In order to effectively detect and evaluate the effect of coal preparation,this paper proposed a mineral components recognition algorithm based on deep learning to classify and locate various mineral components in coal sorting products.Firstly,by analyzing the characteristics of backscattered images of mineral components collected by scanning electron microscopy,a Faster R-CNN object detection algorithm based on deep learning was used to identify and locate the mineral components.Aiming at the factors that interfere with the recognition of minerals such as the holes,fissures,attached tiny particles and the insignificant differences in the mineral characteristics of certain types of minerals in the backscatter image,the attention mechanism was used to improve the feature extraction network in Faster R-CNN for promoting extraction of important characteristics related to various minerals and suppressing the influence of irrelevant interference features.The analysis of experimental results by combining scanning electron microscopy backscatter imaging technology and energy spectrum element analysis technology showed that the algorithm proposed in this paper could accurately and effectively detect each mineral component in coal,and could effectively evaluate the effect of coal preparation.