The Study Of Fine Coal Floatation’s Cleaned Ash Detection Based On Image Processing

In the processing of fine coal flotation, the foam on the mineralized foam layer carries a lot of information of flotation which can reflect the flotation conditions and give an important guiding for flotation operation. At present, the bottleneck of coal slime flotation processing optimization is the accurate description of foam characteristics, the extraction of effective feature parameters, and the establishment of relationship between characteristic parameters and the flotation index, which is also the contents and difficulties that will be discussed in this paper.To study how to use the machine vision technology to extract useful information from flotation foam, in order to provide a reference for the flotation operation, our research team went to a Coal Preparation Plant which belongs to the Xishan Coal Electricity Group and conducted numbers of field trials,haven got a lot of first-hand fine coal flotation foam images, as well as foam coal samples. Based on that, we use a variety of techniques processing each foam image and extract a series of characteristic parameters. After that, we conducted a correlation analysis between these characteristic parameters and cleaned ash. And then, five characteristic parameters which are of highly correlated with cleaned ash are selected, and we conducted the modeling analyses of these experimental data by using multiple linear regression analysis method、the BP neural network analysis method and the method of SVMR. The work is important for achieving real-time optimization of coal flotation process manufacturing operations.In summary, the main research work of this paper can be concluded as follows:(1) Considering the coal slime flotation foam images have a large number of noise disturbance, common image de-noising method was studied, and on this basis, the treatment effect of the average filtering and median filtering method was compared, the results show that the processing of median filtering method is better than the average filtering method.This article also puts forward image enhancement method of our own, and we compared it with histogram equalization enhancement method and wavelet enhancement method, and the effect of enhanced results show that our method for coal slime foam image enhancement has better effect.(2)Studied a variety of image segmentation methods, on this basis, we partitioned the image through the watershed segmentation after preprocessing, from the perspective of the result of segmentation, the segmentation method which was adopted by this paper has achieved a good segmentation result on the whole, which solved the general method of big bubble type excessive segmentation of the image as well as to the problem of small bubble type under segmentation of the image to a certain extent. After the foam images were partitioned, we can calculate the average pixel size of the foam by calculating the number of connected regions, and coupled with the camera’s zoom effects, we can calculate the actual average size of the foam.(3)Based on gray level co-occurrence matrix,we extract the energy, entropy and inertia moment, and based on image gray histogram,we extract the gray average and variance, smoothness, skew, energy and entropy, together with the average foam size that extracted before, there were 10 characteristic parameters. And then the correlation analysis was conducted between the ten characteristic parameters and the cleaned ash, we found that there is a significant relationship between the foam size and the cleaned ash. And there is a high correlation between the cleans ash and the grayscale average, variance, smoothness and skew which were extracted from gray histogram.On the other hand, there exists a moderately correlation between the cleans ash and the entropy extracted from the gray-level histogram and another entropy extracted from gray level co-occurrence matrix,and the relationship between the cleans ash and the energy extracted from gray histogram was low, the relationship between the cleaned coal ash and the energy and inertia extracted from gray level co-occurrence matrix was very weak, we can think they are not relevant.(4) We selected the average foam size and the grayscale average, variance,smoothness and skew extracted from gray histogram as the cleaned coal ash prediction model’s inputs. We using multiple linear regression analysis method, the BP neural network analysis method and the method of SVMR modeling and analysis of these experimental data one by one. The results show that multivariate linear regression model has some validity, but has less accuracy;neural network model for prediction of measuring error due to the number of hidden layer nodes, as long as you control the number of nodes in hidden layer, cleaned coal ash prediction model established by using the neural network can completely has higher prediction accuracy; SVMR model is better than that of multivariate linear regression model but not as good as the neural network model in the aspect of prediction accuracy.(5) Studied the development function of MATLAB interface and use MATLAB software development kit visual interface (GUIDE) developed a control interface of foam image processing system.