Research On Target Recognition And Localization Method Of Coal And Gangue Based On Machine Vision

Raw coal washing is of great significance for clean and efficient utilization of coal.Due to the limitation of washing process,the gangue with large particle size needs to be discharged before washing.The traditional manual gangue discharge operation has low production efficiency,high labor intensity,and the health of workers can not be guaranteed.In order to realize automatic and intelligent gangue discharge,this paper studies the recognition method of coal and gangue based on machine vision,designs and implements an automatic recognition system of gangue.In this paper,the image processing method of coal and gangue is studied,and a complete image processing flow is formed.In the research,a variety of image denoising and sharpening methods are compared;the coal gangue image in the transportation environment of conveyor belt is studied for image segmentation;the image splicing method and target location in real-time processing are studied.The results show that bilateral filtering and Laplacian sharpening have better denoising and sharpening effects on coal and gangue images,and the mean square deviation,peak signal-to-noise ratio and structural similarity of the evaluation indexes of bilateral filtering reach 0.4845,51.2777 and 0.9986 respectively;OTSU algorithm can achieve better segmentation effect,while background subtraction segmentation is not suitable for the current conveyor belt environment;a set of image reality is designed The process of time splicing and processing realizes the target location and position prediction in the image.The image features of coal and gangue are studied.The gray mean,peak gray,gray variance and gray entropy are extracted as gray features.The single factor experiment and multi factor interactive experiment are designed to study the influence of environment on the gray mean of coal gangue.The contrast,energy,entropy and contrast moment are extracted as texture features based on gray co-occurrence matrix.The experiment shows that the gray mean value and peak gray value have good discrimination in the same environment,and can be used as the classification standard of coal and gangue;6-36 W light makes the gray mean value fluctuate about 2,but makes the peak gray value fluctuate seriously,the increase of humidity makes the two decrease greatly,and the mean value and the coal powder amount show a linear inverse relationship;in the multi factor experiment,the humidity and dust are the same The gray mean has interaction,which makes the gray mean decrease about 10,and the decline rate of gangue and coal is different,so the environmental variables should be controlled in practical application.Based on machine learning method,the recognition of coal gangue is studied.The support vector machine classifier is trained with gray feature,texture feature and gray texture combination feature as input,and the key parameters are optimized by particle swarm optimization algorithm.A total of 420 pieces of coal and gangue from Xishan,Shenmu and Inner Mongolia were collected as research data.When 180 groups of coal gangue data were used as input,the optimal penalty parameters of particle swarm optimization were 50,16.95,50,and the optimal width parameters were 31.13,20.42,and 10.02.Finally,the remaining 240 groups of coal and gangue data were divided Class test,the recognition rate is 95.83%,72.92% and 93.75% respectively when the gray feature,texture feature and combination feature are used as input.Based on the image processing,feature extraction,machine learning and other methods studied,and with Microsoft Visual Studio 2017 as the development tool,Open CV visual library is called to process the image,and a set of real-time coal and gangue image acquisition and recognition system is developed in combination with the SDK provided by Teledyne DALSA,which realizes the real-time recognition of coal gangue on the conveyor belt.