Research On Feature Extraction Of Lead-Zinc Ore Flotation Surface

As an optimal alternative of the manual monitoring method,the flotation monitoring based on machine vision and image processing is able to improve the utilization rate of mineral resources at a low cost.Due to complex and diverse characteristics of the flotation surface,how to accurately obtain the visual characteristics of the froth surface in real time is a difficult task.In this thesis,we focus on the visual features extraction of lead-zinc flotation surface based bubble morphological features extraction and froth dynamic features extraction.Our work mainly includes bubble image segmentation,bubble velocity extraction and froth stability characterization and measurement algorithms.The main contributions are shown as follows:1)A method of the bubble edge extraction is proposed based on improved local gray minimum detection(ILGMD).First,a classification algorithm based on Harris corner detection is designed to classify bubble images.Second,the ILGMD algorithm is presented to complete the extraction of the bubble edge candidate points.Finally,the bubble boundary is obtained by post-processing.Experimental results show that ILGMD algorithm has better segmentation result for the images of non-large bubbles:The average bubble segmentation efficiency(ABSE)is 81.4%,the average bubble segmentation accuracy rate(ABSAR)is 89.1%,and the average bubble edges extraction accuracy(ABEEA)is 89.3%;they are improved by 4.5%,3.3%,and 7.2%respectively,compared to the data of all types of bubble images.The algorithm has low complexity and higher degree of automation with no requirement on the adjustment of parameters.2)An improved Watershed(IW)segmentation algorithm of bubble image is proposed with non-subsampled contourlet transform(NSCT)(called NSCT-IW).First,a method of image scale reduction is designed to reduce the processing time and the noise.Second,NSCT is used to decompose the froth image to obtain the low frequency sub-band image,while the high frequency sub-band images are used to design an adaptive fractional differential algorithm to enhance the weak edges of bubbles in the low frequency sub-band image.Finally,the improved Watershed algorithm is applied to image segmentation.Experimental results show that the new NSCT-IW algorithm overcomes the over-segmentation caused by the noise and the black hole areas of large bubbles,as well as the "double edges" problem of large bubbles.The ABSE is 81.9%,the ABSAR is 90.8%,and the ABEEA is 93.1%,which are improved by 2.3%to 16.9%,1.4%to 5.1%,and 1.7%to 6.8%respectively,compared to other algorithms.3)To solve the problem of accurate description on complex bubble movement,we present a bubble translational velocity extraction based on bubble tracking and phase correlation.First of all,the original froth image is reduced by scale.Then the bubble highlight area image is extracted for phase correlation.Then the motion estimation of a bubble is obtained by the motion tracking of its highlight area,which reduces the complexity of motion estimation.Finally,extended Kalman filter(EKF)algorithm is used to optimize the obtained bubble velocity.We also design a block extended phase correlation method to obtain the detailed motion information of bubbles in the sub-block.Experimental results show that the proposed methods accurately obtains the translational velocity of each frame and motion information of bubbles in each sub-block,compared to block matching and optical flow.For the size of 720×480 images,the average processing time of the algorithm based on the bubble tracking and phase correlation is 52.7 milliseconds per frame.The time is reduced by 93.2%compared to the method of phase correlation carried on the original froth image directly.4)We also study the characterization and measuring the stability of froth based on the detection of the feature change of bubble regions.Whether the bubble bursts and merges directly reflects the degree of froth stability.Therefore,six regional variation characteristic parameters of two adjacent frames caused by the two events happen before and after are defined and used to characterize the froth stability.By transforming the froth stability measurement problem into the classification of froth stability,the support vector machine(SVM)is used to train and design the stability classifier to complete the characterization and measurement of the froth stability.Experimental results demonstrate the feasibility of the new method.At last,the feature extraction algorithms proposed in the thesis are applied in a lead-zinc ore flotation plant to extract the data of bubble morphological features,bubble translational velocity and froth stability.Experiments in industrial environment and the analyses of some field experts indicate that the extracted results are effective,and the proposed algorithms are feasible.In summary,this thesis researches on the extraction of bubble morphological features and froth dynamic features of lead-zinc froth images.The algorithm improves the segmentation accuracy of large bubble.The algorithms of dynamic feature extraction efficiently solve the estimation of bubble translational velocity and the measurement of froth stability.Our research is of significance for the realization of the on-line monitoring and optimization control system of flotation.