Flotation Froth-grade Prediction Model Based On Image Processing And BP Neural Network

In the flotation operation,due to the interference of water in the pulp and the limitation of the current detection technology,it is difficult to detect the grade of copper concentrate products in real time and accurately.Domestic concentrators generally obtain grade information based on the experience of operators or use fluorescence analysis technology to obtain grade information in real time,but there is still a lot of room for improvement in its accuracy.In the past half century,detection methods based on machine vision have developed rapidly.Because of their advantages of non-invasiveness,low cost,immediacy and convenience,they have been put into use in more and more industrial fields.In this study,image processing technology and BP Neural Network algorithm were used to establish a color,texture,size statistics and velocity feature extraction model for a copper-zinc-tin polymetallic symbiotic ore in Yunnan preferential flotation copper ore process,and based on the extracted foam image features.Concentrate grade prediction model.The research uses self-built shooting equipment as image acquisition equipment,sampling test results as real grade information,and real-time detection results of fluorescence analysis as a control.Finally,119 original data sets containing foam images,foam flow videos,and main element grade information are obtained.The foam images and foam motion videos in the dataset are respectively passed through the established color feature models based on RGB(Red-Green-Blue)color space and HSV(Hue-Saturation-Value)color space,based on Gray Level Co-occurrence Matrix(GLCM)and Color Co-occurrence Matrix(Color Co-occurrence Matrix).occurrence Matrix,CCM)texture feature model,size statistical feature model based on watershed algorithm,and velocity feature extraction model based on SIFT algorithm.The extracted feature information was screened by the correlation coefficient,and finally 9 features were established as the9-dimensional feature input vector,the Zn,Cu and Mg O grade data were used as the 3-dimensional output vector,and the BP neural network grade prediction model with 3 hidden layers was established.The data set was randomly divided into 100 training sets and 19 testing sets.The Cu prediction results were similar to the fluorescence analysis results,the accuracy of the Zn prediction exceeded that of the fluorescence analysis results,and the Mg O predictions could be made more accurately.Because of the model prediction results are all better than the real-time detection results of fluorescence analysis,and the growth of the neural network model and the wide prediction range,it can be considered that the performance of the copper concentrate grade prediction model based on BP neural network is better than that of fluorescence analysis.Besides,it can be used as a substitute of fluorescence analysis.