Research On Photomicrograph Analysis Of Coal Based On Machine Learning

Coal remains the primary energy in Chinese energy system.Due to the fact that China is rich in coal,short of oil and poor in gas,it will be difficult to change the energy structure in the short term.In addition,under pressure from environmental pollution and energy depletion,clean and efficient utilization of coal becomes a research hotspot,which received much considerable attention in both academia and industry.There are more and more research works on the analysis and separation of coal macerals.The maceral composition is closely related to the physical,chemical and technological properties of coal,which directly affects the quality of gasification,liquefaction and coking.In addition,it is an important basis for the analysis of coal genesis and geological condition of the coal seams.Therefore,it is highly desired to analyze the maceral components in coal.However,the present maceral analysis based on photomicrographs has many concerns to address,such as strong subjectivity,time consuming,labor intensive and high demand of expertise,which seriously restricts the application of coal petrology in various fields.In this thesis,based on the latest research of machine learning and image processing,the research is conducted in the following aspects: image quality improvement for coal photomicrographs;intelligent identification of macerals and estimation of the mean maximum vitrinite reflectance;semantic segmentation of coal photomicrographs based on deep learning.The details are as follows.For low-contrast and blurred photomicrographs,the gamma correction is introduced to improve the visual effect and enhance the contrast of images effectively,which is also very helpful for subsequent image segmentation and analysis.To address the problem of scratches in images caused by grinding and polishing during the preparation of polished grain mounts,a scratch detection and image inpainting method based on Hough transform and Criminisi algorithm is proposed to repair the scratches.These treatments can effectively improve the quality of photomicrographs and lay the foundation for subsequent analysis.Considering the problem that the traditional maceral identification methods are highly subjective and require strong expertise on coal petrology,a novel maceral identification strategy based on image segmentation and classification is proposed.Given the complex and heterogeneous nature of coal,a two-level coarse-to-fine clustering method based on K-means is employed to divide photomicrographs into a sequence of regions with similar attributes(i.e.,binder,vitrinite,liptinite and inertinite).Furthermore,comprehensive features along with random forest are utilized to automatically classify binder and seven types of maceral components,including vitrinite,fusinite,semifusinite,cutinite,sporinite,inertodetrinite and micrinite.Experimental results show that the proposed method achieves the state-of-the-art accuracy of 90.44% and serves as the baseline for future research on maceral analysis.In addition,to support the decisions of petrologists during maceral analysis,we developed a standalone software,which is freely available at https:/github.com/Guyoo Gu/MISC-Master.The vitrinite reflectance measurement process is cumbersome,time-consuming and labor-intensive,and requires high stability of the measurement environment.To address these concerns,a method for estimating the mean maximum vitrinite reflectance of coal based on machine learning is proposed.Considering the complex nature of coal,adaptive K-means clustering is firstly employed to automatically detect the number of clusters(i.e.,maceral groups)in photomicrographs.Furthermore,comprehensive features along with support vector machine are utilized to intelligently identify the regions with vitrinite.The largest region with vitrinite in each photomicrograph is gridded for further regression analysis.The experimental results show that the proposed method can accurately and quickly determine the mean maximum vitrinite reflectance.The predicted values have strong relationship with the reference values,which verifies the effectiveness of the proposed method.In addition,in order to assist the analysis of vitrinite reflectance for petrologists,we released the first non-commercial standalone software for estimating the mean maximum vitrinite reflectance which is freely available at https://github.com/GuyooGu/MMVRML.In view of poor applicability and robustness of traditional image segmentation algorithms,image semantic segmentation algorithm based on deep learning is firstly introduced into coal photomicrographs segmentation.These algorithms can automatically learn appropriate features for segmentation and achieve an end-to-end learning.Four semantic segmentation models based on deep learning are evaluated,and the experimental results show that semantic segmentation algorithms based on Unet series is more suitable for coal photomicrographs.The segmentation results are closer to the manual segmentations,which verify the effectiveness of deep learning based semantic segmentation algorithm in the task of coal photomicrographs analysis.In this study,we integrate the knowledge from many disciplines,including coal petrology,image processing and machine learning.This research results is able to enrich the relevant theories and applications of image processing and machine learning in coal photomicrographs analysis.In addition,the proposed methods can significantly improve the efficiency and accuracy of automatic analysis of coal macerals and promote the application of coal petrology in various fields.The research results of this thesis have important theoretical and practical value.There are 69 figures,15 tables and 145 references in this thesis.