The Study On Image Recognition Of Coal And Gangue Boundary Signatures Based On The Fractional Calculus

Coal is an important resource in China and its reserves are very abundant.Gangue is a type of gray-black rock with low calorific contents and is a companion to coal mining.In order to improve the quality of coal,it is necessary to remove the gangue from the raw coal.With the constant exploitation of coal mines,the produced gangue is also increasing,and the heaped gangue has taken up a large amount of land resources.At the same time,the piled-up gangue has caused pollution to the surrounding environment and has become an urgent problem to be solved.Coal that has been directly mined without any processing is called raw coal in the production of coal.Nowadays,gangue is backfilled to the working surface,made into bricks or paved on roads after sorting the raw coal upground or underground the coal mines,which can reduce environmental pollution,reduce transportation costs,and increase coal production efficiency.Therefore,it is an ideal method for handling gangue.In order to make better use of gangue,coal and coal should be sorted out at the lowest cost.The automatic sorting method of coal gangue is divided into wet method and dry method.The wet method is the main technique for the sorting of coal and gangue in China,such as the dynamic sieve jigging method and the heavy medium method selection method.The wet method selection requires the transportation of coal and gangue into a liquid medium.In addition,the investment is huge and it needs a large amount of water resources.Therefore,it is not suitable for water shortage areas in the west of China.Dry process selection technique does not use water and the investment is small,such as dualenergy gamma ray separation method,X-ray separation methord,image recognition method,etc.However,the radiation-based selection method usually requires high voltage power supply,radiation protection devices and radioactive sources.The management requires strict requirements and it is difficult to use underground.At present,many coal mines in China still use manual selection methods to seperate gangue from the raw coal,where workers are standing on both sides of the belt conveyor to pick out coal or gangue.The craftsmanship of manual selection is very primitive and laborious with low work efficiency,and the noise and dust are harmful to workers.Therefore,the need to use automatic selection technology to replace manual selection is very urgent.The image recognition method aims at identifying the characteristics of the surface and shape of coal and gangue,using computer vision and pattern recognition technology to quickly determine whether the target is coal or gangue.At the same time,the image recognition method does not need water.It has a small size,low cost,no radiation and simple structure,strong transferability.Therefore,the use of image recognition method to replace the manual seperation has a promising application prospects.The focus of the coal-gangue separation method based on image recognition is the extraction of the characteristics of coal and gangue.In the current literature published nationwide,researchers mainly study and identify the statistical features such as the mean value and variance of the coal gangue image.The information such as the mean and variance of the image histogram is used as the characteristic vector.However,due to the different sizes of coal and gangue,the selection of grayscale and texture features usually requires intercepting part of the image as a recognition sample.This results in the loss of information and influence.In addition,after the gangue is attached to the coal powder,its grayscale and texture characteristics are very similar to that of coal.The recognition rate is poor with this method.The outline characteristics of coal and gangue belong to physical properties and are not susceptible to interference.Because the coal and gangue have different characteristics such as density and hardness,the edge contours are different after several times of collisions during the transportation.The research on coal and gangue outline curves is the focus of this paper.Fractional calculus is a generalization of the conventional integer order calculus,providing new thoughts and perspectives.In this paper,the research of coal-gangue boundary signature image recognition based on the fractional calculus is studied,using the boundary signature method to describe the shape of the coal and gangue curves,to obtain a one-dimensional function expression from the center of mass to the boundary,and then to extract the geometric characteristics of the coal-gangue profile curve.Through the modeling and analysis of the fractional-order stochastic process of coal and gangue boundary signature series,it is pointed out that the coal-gangue boundary signature series has multi-fractal characteristics such as non-stationary,non-Gaussian,self-similar,etc.Multifractaldetrended fluctuation analysis(MFDFA)is first discussed and then the multifractal spectrum width ?? is chosen as the geometric characteristics of the coal gangue outline curves;then an improved MFDFA algorithm is proposed to obtain the multifractal spectrum of the coal gangue boundary signatures.Finally,the recognition rate of different model methods is adopted.The result shows that the introduction of the coal gangue's geometric features is helpful to improve the recognition rate of the coal gangue image.In view of the above problems,this paper uses some theoretical knowledge such as mineral processing,optics,advanced mathematics,statistics,image processing,signal analysis,fractal and multifractal analyisi,time series analysis and computer science to focus on the extraction of the characteristics coal gangue recognition under the current working environment.This paper proposes the use of fractional calculus for the first time to study the boundary signatures of coal and gangue.By establishing a fractional-order stochastic model and combining multifractal analysis methods,the geometric features of the coal gangue profile are improved and extracted.Based on the existing research results,the grayscale and texture features of the image are obtained.Through the analysis of the data,calculation and experimental verification,the identification of the coal gangue image is completed.This article mainly performs the following aspects of the work:(1)Using boundary signature to represent the outline curves of coal and gangue.Using digital image processing technology,such as image cropping,denoising,image restoration and enhanced morphological processing,the principle and method of extracting image outline curve by boundary signature method are discussed.The edges of the coal and gangue are identified and spread along the center of mass in polar coordinates.The one-dimensional expression of the centroid-to-boundary distance is obtained.It is also pointed out that the series of coal-gangue outline curves is nonstationary,non-Gaussian,and self-similar.(2)Fractional modeling of coal and gangue profile curves.According to the constant order fractional Brownian motion,the traditional Gaussian distribution model is extended to a non-Gaussian ? stable distribution stochastic model to describe the distribution of complex processes.Aiming at modeling the series of coalgangue outline curves,an auto-regressive fractional autoregressive moving average(ARFIMA)model with long memory characteristics was proposed.The relationship between the Hurst parameter,the fractional order difference factor and the characteristic distribution of the ?-stable distribution is derived.The modeling and analysis of the fractional-order stochastic process are performed.The existing Hurst parameter estimation method and the robustness of each estimator are comprehensively evaluated and improved through data simulation and measured data.It provides theoretical basis and analysis tools for the analysis of profile features of coal gangue.Through modeling and analysis of the coal-gangue boundary signatures,the fitting result of the ARFIMA model is better than that of the ARIMA model.At the same time,according to the fitting failure of the gangue boundary signature,it is pointed out that the outline series of the gangue has multi-fractal characteristics,and then the variable order is introduced with multifractional process and multifractal analysis method.(3)Improving multifractal analysis methods to determine the geometric features of coal and gangue curves.In order to analyze the complex signals with local self-similarity characteristics and local memory,simulation examples and parameter estimations of multi-fractional processes and fractional-order stochastic processes of local H?lder variables are presented.MFDFA method is used in the paper to represent the meaning of white noise sequences,single fractal sequences,and multifractal simulation series at different scales and orders.This method not only effectively eliminates interference trend terms,measures the fluctuation scale of nonlinear time series,but also can accurately estimates the multifractal spectrum.For multifractal series with different structures,the effect of correlation and distribution is eliminated through shuffling and surroggating data.Thus,the value of multifractal spectrum width ?? is modified,making the singular spectral width of coal and gangue outline curve series tend to be reasonable.As a result,the separation threshold of the local singular spectral width of the coal gangue curves after the revised algorithm is around 0.48.(4)Combining geometric features with grayscale and texture features,and evaluating the recognition rate of coal gangue images.The proposed fractional model and multifractal analysis method are applied to the geometrical boundary signatures of coal gangue images.The geometric features of the coal gangue outline curves are extracted.The mean,consistency,multifractal spectrum width ?? are selected as the grayscale features,texture features,and geometric features of coal gangue images respectively.Then different feature vectors and pattern recognition methods are compared.The 500 coal gangue image samples collected in the coal preparation workshop are used for training and testing.The experimental results show that the recognition rate of coal gangue can reach 97.5% under the kNN identification algorithm with the geometric features combined with the gray features and texture features.It indicates that introducing the geometric features based the fractional calculus method can improve the recognition rate of coal gangue images.