Froth Image Feature Extraction Based On Wavelet Multi-scale Analysis For Copper Flotation

Abstract:Froth flotation is the most widely-used method in mineral separation process. At present, the states of the surface bubbles in flotation cells are inferred by observing with naked eye, touching with hands, which greatly limits the optimal operation of the froth flotation process and degrades the mineral recovery. Therefore, it is very important to study the morphological features of froth images and realize the optimal control of froth flotation process.Focused on the characteristics of copper flotation bubbles that they flow fast and tend to gather and collapse, visual features such as bubble size, texture et al. are choose to descript the image information of bubbles. Based on the multi-scale property of wavelet transform, methods which combined wavelet transform and other methods are proposed. The support vector machine (SVM) is applied to classification and recognition of froth images. Below are several aspects of our work:Firstly, to avoid the complex segmentation algorithm, a binarazation method is applied to extract size features of froth images. Binary results of the iterative method, OTSU method and minimum error thresholding method are evaluated, and the bubble size features extracted by minimum error thresholding method is supposed to be more accurate.Secondly, Traditional statistical features of froth images extracted by wavelet transform have multi-scale property, but are not able to reflect froth morphology characteristics directly. Bubble size features are easily to be obtained through image binarizing, yet lacking multi-scale property. Aiming at the above problems, a method for feature extraction of froth images based on wavelet multi-scale binarization (WMB) is proposed. Experiment results demonstrate that the froth image classification with SVM has achieved a good effect with the equivalent size feature, so that working conditions of copper flotation can be recognized well.Finally, since statistical features generated from gray level co-occurrence matrix (GLCM) can only describe the texture structural information on the spatial domain of images, which is obtained at a single scale, there is a loss of dependency relationship between texture scales. Wavelet analysis enables decomposition of original images into sub-band images of different frequencies or resolutions. However, wavelet analysis is not suitable for the analysis of structural information of froth texture. A method based on multi-scale gray level co-occurrence matrix (MGLCM) is proposed to extract texture feature of froth images. Experiment results demonstrate that the MGLCM feature has good stability and separability. Off-line classification and On-line recognition based on the MGLCM feature are applied to copper froth flotation and have achieved satisfying results. This paper contained30figures,6tables and65references.