Sulfur Flotation Performance Recognition Based on Hierarchical Classification of Local Dynamic and Static Froth Features

This paper proposes a flotation performance recognition system based on a hierarchical classification of froth images using both local dynamic and static features, which includes a series of functions in image extraction, processing, and classification. Within the integrated system, to identify the abnormal working condition with poor flotation performance (NB it could be significantly different with the dynamic features of the froth in abnormal working condition), it is functioned first with building up local dynamic features of froth image from the information including froth velocity, disorder degree, and burst rate. To enhance the dynamic feature extraction and matching, this system introduces a scale-invariant feature transform method to cope with froth motion and the noise induced by dust and illumination. For the performance subdividing under normal working conditions, bag-of-words (BoW) description is utilized to fill the semantic gap in performance recognition when images are directly described by global image features. Accordingly typical froth status words are extracted to form a froth status glossary so that the froth status words of each patch form the BoW description of an image. A Bayesian probabilistic model is built to establish a froth image classification reference with the BoW description of images as the input. An expectation-maximization algorithm is used for training the model parameters. Data obtained from a real plant are selected to verify the proposed approach. It is noted that the proposed system can reduce the negative effects of image noise, and has high accuracy in flotation performance recognition.