Design Of Intelligent Separation System For Coal And Gangue Based On Machine Vision

Increasing the washing rate of raw coal is an important way to improve the quality of coal,which is also in line with the trend of clean and efficient utilization of coal resources.Before the raw coal is washed,the gangue with larger particle size in the raw coal should be sorted to meet the requirements of subsequent crushing and washing.Traditional coal gangue separation is mainly done by manual,which has the problems of high labor intensity and low separation efficiency.In order to realize the automation of coal gangue sorting,this paper designs a set of intelligent separation system of coal gangue based on machine vision.The system is mainly composed of the coal gangue sorting platform which is responsible for collecting images and dynamic testing,the upper computer gangue sorting software responsible for image recognition and positioning,and the control system controlling the action of the actuator.(1)The paper built a coal gangue sorting platform in the laboratory,selected the belt conveyor with black belt to simulate the actual working conditions of the coal preparation plant,introduced the selection and installation of key equipment such as camera and light source,and collected 716 coal gangue pictures as the data set of subsequent image processing based on the platform.(2)This paper uses a single hidden layer of multi-layer perceptron and deep target detection network to identify coal gangue.When using multi-layer perceptron to identify coal gangue,this paper studied the feature extraction algorithm of coal gangue image in detail,it can be concluded that the coal gangue image features can be directly extracted after Gaussian filtering,background subtraction,gray linear transformation and binary threshold segmentation,and the multi-layer perceptron with 10 hidden nodes was selected as the identification network;when the target detection network was used to identify coal gangue,the feature pyramid in the target detection network was set to three scales,and the shape and size of anchor were determined.In addition,by comparing the average precision of verification set under 10 Io U(intersection over union)thresholds,it was concluded that the classification and positioning effect of target detection network was the best when Io U was 0.8.By comparing the validation sets of the two network models,it can be concluded that the deep learning represented by the target detection network is more suitable for the identification and location of coal gangue under complex working conditions than the shallow learning network represented by the single hidden layer perceptron.(3)According to the data after the coal gangue image recognition and positioning and the hardware characteristics of the sorting platform,the RS485 serial port communication was adopted for data transmission,the information frame of 8 data bits was designed according to the recognition and positioning data after image processing,PLC was selected as the lower computer to receive data,and the control program which can control several sets of actuators separately was designed.(4)In order to realize the function of sorting system,this paper called the image processing program of HALCON based on MFC to complete the preparation of the upper computer coal gangue separation program,and developed the human-computer interaction interface of coal gangue separation on this basis,which integrated the image acquisition,processing and data transmission functions of the whole system.In addition,in order to accurately evaluate the performance of the system,the dynamic test was carried out on the built coal gangue separation platform.The results show that: if there are many coal gangues on the belt conveyor and the particle size difference is large,the multi-layer perceptron method will be unable to segment in the image segmentation,and the accuracy rate of dynamic recognition is 84.5%;while the target detection network method can accurately classify and locate the coal gangue,and the precision and recall rate of dynamic recognition are more than 95%.The coal gangue sorting system composed of the coal gangue sorting platform built by the laboratory,the upper computer coal gangue identification software using the target detection network and the PLC control system can complete the dynamic identification of coal gangue in the case of less coal gangue and more concentrated particle size.The queuing mechanism and actuator of the separation platform should be further optimized to realize automatic separation under actual working conditions.In addition,this study can provide reference for the application of coal gangue separation system in coal preparation plants.