Classification Research Of Coal Flotation Froth Based On Machine Vision

At present,China’s energy structure is facing the challenge of transformation and development,the coal industry as the pillar of national energy,improve the intelligence level of its flotation process is the necessary way to realize the change and upgrade of coal processing industry,but also to promote the green and efficient development of the whole coal industry,most of the coal processing plants are still in the stage of low information level,intelligence is just starting,for the monitoring of the flotation process mainly relies on manual operation,which has the problems of strong lag and low accuracy.With the widespread use of machine vision technology,the flotation process can be discriminated by the image analysis of coal flotation froth,which has become a hot direction for intelligent flotation.This paper firstly designed and built a visual acquisition system for flotation froth images,and designed seven sets of experiments by changing the concentration,aeration and dosing amount,and analyzed the flotation froth images obtained under different working conditions.Then for the problems of low contrast and high noise in the flotation froth images,CLAHE algorithm is used to improve the contrast between light and dark,and for the noise in the images,BM3 D method is used to denoise the froth images,and Laplacian operator is used to sharpen the froth images to improve the quality of the images,and finally a good pre-processing effect is achieved.The final pre-processing result is better.Multi-dimensional features were extracted from the pre-processed froth images: firstly,multi-dimensional color features were extracted in RGB color space,YUV color space,Lab color space and HSI color space;secondly,for the problem that the flotation froth images did not work well with traditional segmentation methods,the labeled data set was trained with the deep learning network U-Net,which finally achieved a good segmentation effect.and extracted accurate morphological feature parameters;next,LBP texture features and GLCM texture features were extracted from the froth image,and the texture information of the froth was considered from multiple angles;the feature point detection was performed between two frames of froth images using ORB corner point detection algorithm,and the accuracy and speed of feature point matching were improved by BEBLID algorithm,and for the existing wrong matching points,the RANSAC algorithm for filtering,based on which the velocity features of the froth images are obtained,and finally 83-dimensional feature vectors are obtained for each flotation froth image sample data.The results showed that the classification accuracy of Res Net50 network reached 95.18%when the SGDM optimizer,learning rate of 0.0007 and 5 iterations were used.The results show that when the SVM constraint scale is 0.0033,the linear kernel function and the one-to-many classification strategy are used,the best classification result is achieved with an accuracy of 97.3%,and the real-time performance of SVM is good.Finally,based on the above study,the flotation froth image analysis software was designed,focusing on the actual needs of the flotation industry,including the interaction between the software and PLC process data,the froth image analysis module and the database operation and maintenance management,etc.The entire software program was written in the Qt Designer interface using the Py Qt5 toolkit,and a simple and clear human-computer interface was designed.It is designed with a simple and easy to operate human-machine interface.