Evaluation Of Flotation Production Condition Based On Image Multi-scale Feature And Extreme Learning Machine

The surface visual characteristics of flotation bubbles are direct indicators of flotation conditions and process indices.In the actual mineral sorting production,the surface visual characteristics of the flotation bubbles rely on artificial observation,and it is strong subjectivity,big error and low efficiency.It is impossible to realize the objective evaluation of the flotation state,which results in the increase of the production costs.With the rapid development of computer,the digital image processing technology is widely applied to the study of flotation automation monitoring.Image processing technologies such as nonsubsampled Contourlet transform and binary are used to analyze the multi-scale characteristics of flotation bubbles,and to establish the linkages between these characteristics parameters and the flotation indexes.The flotation production conditions are identified by the improved kernel extreme learning machine(KELM)model.The main research content and the contributions of this thesis can be summarized as:1.Flotation bubble image preprocessing.In view of the problem that the extraction of flotation bubble is inaccurate,the nonsubsampled Contourlet transform is used to decompose the flotation bubble image in multiple scales to form a low frequency subband and multiple high frequency sub-band images.This thesis focuses on the components of the high and low frequency sub-band image,and adjusts the multi-scale distribution of the flotation bubble image information by changing the parameters of the nonsubsampled Contourlet transform.The low-frequency image is enhanced to improve the brightness and highlight the bright spot information.After the high-frequency image is de-noised,the bright spot edge noise is removed and the bubble edge details are preserved as much as possible.This method can provide more effective information for the subsequent multi-scale feature extraction.2.Multi-scale feature extraction of flotation bubble image.Aiming at the multi-scale distribution of the information in the floatation bubble image,we firstly use the binarization method to extract the highlights from the low-frequency image and the high-frequency scale image,then extract the bubble equivalent size and morphological features from the bright spot image,and extract the texture features such as mean,variance and fractal dimension in the high-frequency image.Then,the multi-scale features are analyzed,which shows that the state recognition of flotation bubbles cannot rely on a single feature,and it needs to combine multiple image features.Finally,by comparing with the traditional features,the results show that the non-subsampled Contourlet multi-scale features proposed in this paper have higher prediction accuracy and stronger ability to track the trend of change.3.The establishment of the evaluation model.Aiming at the problem that the learning parameters of the extreme learning machine can be randomly selected to influence its stability and generalization ability,a model of image classification evaluation based on Kelm is proposed.In order to solve the sensitivity problem of Kelm learning parameters,the improved quantum genetic algorithm is used to optimize the learning parameters of Kelm.The experimental results show that the improved Kelm model has better recognition effect than other models.Therefore,it is more advantageous to apply the new model to the evaluation of flotation condition.4.Evaluation of flotation conditions.In view of the abnormal problem of the flotation bubble image data,the preprocessing operations such as culling and normalization of the data are carried out.The improved KELM evaluation model is established by training the obtained normal data,and the test set is identified and analyzed by using the established model.Finally,the experimental results show that the model proposed in this paper improves the identification rate of flotation conditions.