Image Processing Method And Experimental Research Of Mineral Flotation Based On Deep Learning And Gray Scale Analysis

The intelligentization of mineral flotation is an important issue in the utilization of mineral resources and the flotation industry.Increasing the flotation efficiency can effectively increase the utilization rate of mineral resources,reduce enterprise costs,improve energy efficiency,and reduce environmental pollution.In the research of flotation technology,the intelligent monitoring of flotation froth image is the focus of research in the field of coal flotation.However,in the actual industrial flotation process,ash content and other important indicators related to the quality of flotation have been obtained by manual sampling and testing for a long time,which has serious lag,and only a few periods of ash data can be obtained,and real-time online cannot be achieved.monitor.Therefore,the traditional coal flotation process is difficult to adjust the production situation in time,and it is difficult to ensure that the produced clean coal is qualified and stable.In recent years,with the rapid development of computer image processing technology,machine vision has been widely used in industry,and artificial intelligence and deep learning can become important tools for flotation intelligence.A large number of studies have shown that the visual characteristics of the flotation froth surface contain a lot of information related to flotation conditions and production indicators,such as the correlation between the ash content of coal and the gray level of the froth surface.Although traditional image processing algorithms have achieved some results in this regard,the upper limit is not high.Although the emerging deep learning method theoretically has a very high accuracy rate under the condition of sufficient sample training,but due to the inherent conditions of the flotation industry,it is relatively difficult to collect and label data,and it is difficult to froth a large data set.A common industry problem with reliable public data sets.Aiming at the above difficulties and shortcomings,this thesis relies on coal flotation experiments,researches mineral flotation images with coal as a representative,and uses computer image processing algorithms to study coal flotation froth images.The specific contents include:(1)Propose an improved M-Deep Labv3+ neural network based on Deep Labv3+,and use deep learning algorithms to study the influence of factors such as bubble generation rate on the flotation effect.In the laboratory environment,the XFD-63 frequency conversion speed-regulating single-slot flotation machine is used for coal flotation experiments.Choose to collect videos that include all the processes of flotation operations,select screenshots from the videos according to the time point for research,and use expert consultation with the actual needs of the algorithm to mark.Complete the segmentation of the bubble image.The experimental results show that compared with the traditional contour extraction algorithm and the original neural network,the improved network has higher accuracy.(2)Propose a coal ash change detection method based on flotation froth image.This method can measure the relative ash content by calculating the gray level of the foam.It has strong anti-noise ability and high versatility.It can make up for the shortcomings of low precision of deep learning under small sample conditions and monitor the quality of flotation conditions.(3)Design a set of relative ash change monitoring system that can be used in flotation site.This system can better guide the production situation on site.In addition,the experimental coal flotation foam image data collection process can effectively obtain high-quality available data sets,forming a set of standard process for mineral flotation foam image data collection,laying a foundation for deep learning in subsequent related research applications.In this thesis,there are 44 figures,14 tables and 65 references.