Bauxite Froth Flotation Working Conditions Recognition Based On SVM Multi-class Classification Algorithm

During the mineral flotation process, foam layer can be a direct response to the quality indicators. At present, the control of flotation process is often based on the visual appearance of the froth phase, and to a large extent depends on the experience and ability of a human operator. These types of process are consequently often controlled sub-optimally owing to the inaccuracy and unreliability of manual control. Therefore, researching the recognition of the froth flotation working conditions and applying them into practice are great significance for optimizing flotation process, maximum using resource, reducing resource consume and maintaining enterprise sustainable development.SVM is based on the principle of structural risk minimization and has good generality capabilities. SVM has been widely used in various fields, e.g. pattern recognition and regression estimation. Initially SVM was proposed only for binary classification problems. How to promote binary classification to the multi-class, and effective applicate it in the research of bauxite froth flotation condition recognition, have high theoretical and practical application significance.In the novel one-against-many multi-class classifier, the number of training samples for sub classifier is very huge which leads to a low classify accuracy. As to the one-against-one-algorithm multi-class classification support vector machine, the classification speed is very slow for it needs to create a large number of sub classifier. In this paper, a one-against-two multi-class classifier is proposed; it can significantly improve the classification accuracy and classification speed. In the novel Binary tree decision and Decision Directed Acyclic Graph (DDAG) algorithm, the structure of the classifier varies from the sample collections, so the classification accuracy and classification speed also changes. To avoid this problem, a convex hull binary tree algorithm multi-class classifier is proposed, which generates a binary tree structure that can reach best classifier result. This method ensures the speed and stability of classification. The UCI data sets simulation results show the effectiveness of the proposed algorithm.Finally, in the study of bubble bauxite layer surface relations on the basis of visual characteristics and flotation conditions, bauxite froth flotation working conditions recognition based on SVM multi-class classification algorithm is proposed. Bauxite production field data experiments show that the proposed method can well reflect the working condition of different foam layer on the surface of the difference of visual features, and have very good bubble image classification recognition effect, which can be controlled to achieve the optimization of the flotation process to provide more effective instruction.