Image Segmentation And Morphological Analysis Of Coal Particles Based On Deep Learning

Coal plays an important strategic role in China’s energy consumption.Raw coal is usually inevitably mixed with various minerals in the production process,which leads to the decline of raw coal quality.Its combustion process is very likely to cause environmental pollution.The coal preparation process can effectively remove minerals from coal,which is of great significance to the efficient and clean utilization of coal.In the process of coal separation,the particle size information of pulverized coal and mineral content in coal are closely related to the separation effect.And it is an important parameter index for evaluating coal preparation technology.As the rapid development of deep learning and image acquisition equipment,the application of artificial intelligence technology in the field of coal preparation can accurately obtain the morphological information of pulverized coal particles and the mineral content information in coal,which is conducive to optimize the preparation process and improve the production efficiency of mineral separation.According to the characteristics of images in different modes,corresponding segmentation algorithms are proposed to achieve accurate segmentation of pulverized coal particles,and particle morphology research and mineral content analysis in coal are carried out based on the segmentation results.The main research work of this thesis is as follows:1.Aiming at the problems of under-segmentation,over-segmentation and false segmentation caused by particle adhesion,uneven particle size and irregular shape in SEM-SE images of pulverized coal,this thesis proposes a segmentation algorithm for SEM-SE images of pulverized coal particles based on Mask R-CNN,a two-stage segmentation algorithm.Firstly,in order to improve the segmentation accuracy of the network for small particle targets,attention mechanism is integrated into FPN(Feature Extraction Network)to enhance the channel information and semantic information.Secondly,the branch structure of mask segmentation is redesigned by using skip connections to further promote feature fusion and retain more feature information,so that the network can accurately segment complex particle targets.Experiments on self-built data sets show that the proposed algorithm can effectively segment the SEM-SE image of pulverized coal particles.At the same time,this thesis selects several parameters to characterize the size and shape of the segmented pulverized coal particles.In addition,the functional relationship between particle size and particle size distribution is obtained by data fitting,which proves the effectiveness of particle size analysis method based on image processing.2.To evaluate the effect of coal preparation and give effective feedback,this thesis uses SEM backscattering imaging mode to collect images of coal particles after separation,and proposes a semantic segmentation algorithm based on deep learning to segment the SEM-BSE images of coal particles.In order to solve the problems of similar features,blurred edges and particle adhesion in SEM-BSE images of coal mines,this thesis proposes a coal mine particle image segmentation algorithm based on the semantic segmentation model Swin-UNet.Firstly,a convolutional neural network integrating attention mechanism is designed as an auxiliary encoder to capture local features,which is used as the information supplement of the main encoder,so that the network can capture global dependencies and local details at the same time.On the one hand,it can accurately locate small particle targets and reduce the problem of missing segmentation.On the other hand,it can improve the recognition accuracy of coal and minerals by network.In addition,this thesis adopts the joint training strategy of dice loss function and focal loss function to promote the network learning of particle edge features and effectively improve the segmentation accuracy of particles.Experiments show that the proposed algorithm is more robust than other mainstream algorithms in the SEM-BSE image set of coal particles.At the same time,this thesis selects appropriate parameters to count the content of coal and mineral particles in the segmentation results,which can be used as the evaluation basis of coal preparation effect.