Research On Adaptive Classification And Recognition Algorithm Of Coal Macerals Based On Multiple Features

The classification and recognition of coal macerals determine the coking ratio and process properties of coal,and it has been a hot research direction in the field of energy and chemical engineering.Automatic identification of coal macerals can improve the speed and efficiency of coal rock classification,and its analysis results are more objective and reliable.At present,the automatic identification methods of coal macerals are mainly based on image processing methods.Although great progress has been made,there are still many problems to be solved in the application process.This subject mainly solves the problems of low recognition accuracy and low efficiency of coal macerals identification.By preprocessing,an adaptive correction and automatic identification of coal macerals,the overall quality,sharpness and contrast of grayscale images are improved,and complete the classification and recognition of coal macerals accurately.The rationality of the proposed algorithm was experimentally verified and analyzed during research.Based on the above,the main research work is as follows:(1)Location and restoration of scratches and false boundary areas in the gray image of coal macerals.In the process of making and collecting coal samples,there will inevitably be some scratches and false boundary areas.These areas occupy a certain number of pixels,which will interfere with the identification accuracy of coal macerals,and must be processed in the pre-processing stage.Based on characteristics of scratches,this paper uses Hough Transform to locate the scratches,and then uses the neighborhood mean replacement method to repair the scratches of coal,until all scratches are removed.In addition,the Prewitt operator with 3 × 3 template is used to detect the edge and locate the false boundary area in the grayscale image.It is corrected to the background group to avoid the phenomenon of false classification and segmentation caused by the false boundary.(2)An adaptive truncated gamma correction algorithm is proposed to correct gray image of coal macerals.This paper aims at the problems of degradation,distortion and blur caused by light changes in grayscale images.This method is mainly divided into adaptive stage and truncation stage.In the adaptive stage,firstly,the probability distribution of gray value and the variable gamma parameter are used to improve the contrast of low-quality image.This step can enhance the brightness and contrast of the dark area of the image,but it is not effective in the bright area.In order to overcome this defect,the curve of gamma parameter is truncated by a reasonable threshold.At this time,the contrast of the image can be changed without losing the details of the bright area.The experimental results show that the adaptive truncated gamma correction algorithm can effectively suppress image noise,enhance the contrast,brightness and clarity of the images.(3)A classification algorithm of coal macerals based on spatial information and double decomposition Gaussian mixture model is presented.The traditional Gaussian mixture model has the problem of poor classification accuracy due to inaccurate fitting results and failure to consider the spatial relationship between image pixels.Based on this,the model in this paper is mainly improved in two steps.The first step is to double decompose the original Gaussian mixture model.Firstly,the model is decomposed into K Gaussian models,and one of these K models is decomposed into S Gaussian distribution,which is conducive to improving the fitting accuracy of the model.The second step is to integrate spatial information.The spatial information constraints are added to the probability distribution function and the posterior probability in the model to enhance the information association and anti-noise between pixels.After a variety of sample verification and analysis,compared with the traditional Gaussian mixture model method,the improved KMeans method,and the histogram peak fitting method,the recognition result of the algorithm in this paper is the closest to that of the standard data set,and 30 coal samples are selected.Compared with the standard data set,average difference between the recognition result of the vitrinite,inertinite,and semi-vitrinite are only 2.21%,2.5%,and 2.57%,which means that the errors are generally lower than the other three recognition methods.