Online Monitoring of Flotation Froth Bubble-Size Distributions via Multiscale Deblurring and Multistage Jumping Feature-Fused Full Convolutional Networks

This article proposes an online bubble size distribution (BSD) monitoring scheme by incorporating a multiscale-deblurring full convolutional network (FulConNet) (MsD) and a multistage jumping feature-fused FulConNet (MsJ), having the potential of online identification of the health state of the flotation process operations. MsD can restore the blurry froth images from any complex foggy and motion-blurred scene due to air-water fogs, camera vibrations, and high-speed froth flows. MsJ is proposed to segment accurately various froth images with the fully occupied and closely adhesive fragile bubbles, involving multiple residual groups and multiple jumping feature layers to delineate the bubbles of various sizes adaptively. The Weibull distribution behavior of the left-skewed BSD is demonstrated by the sequential fragmentation theory, whose parameters can be used to identify the flotation state. Extensive comparative experiments on a real copper-mine flotation process demonstrate that the proposed method performs favorably against the state-of-the-art froth-image-segmentation approaches, and the Weibull distribution model can effectively characterize the underlying left-skewed BSD behavior, which is an effective indicator for the online flotation-state identification or health evaluation.