Prediction model of iron reverse flotation tailings grade based on multi-feature fusion

Abstract

During the production of iron in reverse flotation, obtaining real-time froth grade values helps to adjust the reagent regime to ensure the quality of the beneficiated product and save production costs. In predicting the froth grade, it is necessary to segment the froth image to extract bubble size features. In this study, an improved U-net algorithm with a focus mechanism is proposed to effectively avoid under-segmentation and over-segmentation and to extract bubble size features. Through the analysis, the bubble color features that are beneficial to the rank prediction are identified. A multi-feature fusion deep neural network model based on Resnet is established to predict the bubble grade values. The test results indicate that with sufficient accuracy, the model can be employed in reverse flotation production.