Intelligent Identification Methods Of Coal Macerals And Applications

Intelligent identification method of coal macerals was studied,to reduce the subjectivity of maceral identification,and improve the accuracy and efficiency.727 commercial coal samples were systematically collected from 25 provinces in China,and the petrographic characteristics of different coal-forming ages,coal-accumulating areas,and coal types were analyzed.The average maceral content based on reserves weight was put forward to obtain the general characteristics of maceral content in Chinese commercial coals.The petrographic images of 24 typical coal samples were collected by a CCD camera,and a dataset of maceral characteristics of bituminous coal was established,covering the main coal-accumulating areas and coal-forming ages.Three optically isotropic standard samples,sapphire,yttrium aluminum garnet(YAG)and gadolinium gallium garnet(GGG),were used to study the influence of uneven lighting of microscope on the characteristics of petrographic images as the field scope of microscope is expanded.The calibration methods based on image processing technologies were then proposed.Image preprocessing technologies of surface unevenness of petrographic blocks,defocusing classification method by image definition,and separation of touching particles were analyzed.The identification characteristics of coal macerals under reflected light with oil immersion were summarized.The characteristic parameters of gray level,morphological feature and texture of 18 kinds of macerals were studied by gray level co-occurrence matrix.Two machine learning classification algorithms,k-nearest neighbor(KNN)and support vector machine(SVM),were used to classify the interactive features of gray and texture in 739 petrographic images respectively.Three semantic segmentation models based on deep learning,U-Net,Seg Net and Deep Lab V3+,were used to make a pixel level identification of maceral groups in Chinese bituminous coals.Pixel accuracy(PA),intersection on union ratio(Io U)and BFScore were introduced to evaluate the recognition effect of the models.And comparisons were conducted with measured samples.The heating stage microscope,fluorescence analysis and Micro-FTIR were used to study the in-situ pyrolysis of macerals.The characteristics of macerals were extracted and analyzed by the intelligent recognition method under heating stage microscope and fluorescence microscope.The main conclusions were as follows:The petrographic compositions were quite different in different coal-forming ages,coal-accumulating areas and coal types.The weighted average contents of vitrinite,inertinite,liptinite were 61.3%,36.5%,and 2.2%,respectively.Relatively high content of inertinite and low content of liptinite were found of Chinese commercial coals.The reflected light on the surface of the blocks were formed because of the uneven lighting of the light source of the petrographic microscope,then shadow was formed in the digital image,as the field scope of microscope is expanded.The shadow image caused by uneven lighting could be calibrated by lowpass filter and top-hat transform,as shown by the analyses of standard samples and coal blocks.The top-hat transform was preferred in petrographic image calibration.The blur in petrographic images and shadow on particle edges were caused by the relief differences of macerals.The concave point matching method was used to separate the touching particles,which benefited the identification of coal macerals.For the first time,the characteristics of gray scale and texture of 18 kinds of macerals,i.e.telinite,desmoclllinite,telollinite,corpocollinite,gelocollinite,fusinite,semifusinite,macrinite,macrinite,funginite,secretinite,sporinite,cutinite,resinite,suberinite,clay mineral,pyrite and calcite,were systemically studied in maceral level.More than 60% recognition accuracy was achieved by the support vector machine(SVM)algorithm based on gray and texture features for the 18 kinds of macerals.The loss curves and accuracy curves of U-Net,Seg Net and Deep Lab V3+ models converged rapidly after iteration.The average pixel accuracy(PA)of the three models was 73%,73% and 92%,respectively.Both the Io U and BFscore of Deep Lab V3+models were significantly higher than those of U-Net model and Seg Net model.The Deep Lab V3+ model was the preferred method for maceral group identification considering the training efficiency and segmentation accuracy.The maceral group compositions determined by Deep Lab V3+ model achieved similar results as the ISO method,and met the accuracy requirements of quantitative analysis of coal maceral groups,which indicated that the macerals could be identified at the pixel level by the semantic segmentation method based on deep learning.The characteristic parameters of macerals in images under heating stage microscope and fluorescence microscope could be effectively extracted by the intelligent recognition method.The softening temperature of the liptinite(including sapropelic groundmass)was about 350 ～ 370 ℃ while that of vitrinite was about 410 ～420 ℃ as shown by the image analysis.The pyrolysis process of vitrinite went through the stages of edge shrinking,pore formation,surface softening,formation of liquid phase,and solidification.Only slight morphological changes were observed in semi-vitrinite,while no changes were observed in inertinite.The fluorescent characteristics of liptinite components in coal changed regularly.As the pyrolysis temperature increased,the relative fluorescence intensities of liptinite components decreased.The fluorescence characteristics of resinite and suberinite changed remarkably in 280～320℃ while those of sporinite,cutinite,bituminite A changed remarkably in 320～360℃.The fluorescence characteristic of alginite began to change at 280℃,and lasting to 400℃.The fluorescence characteristic of bituminite B changed in 320～360℃.