FS-DeblurGAN: a spatiotemporal deblurring method for zinc froth flotation

Wenhui Xiao, Zhaohui Tang, Jin Luo, Jinping Liu

Abstract: Flotation froth image deblurring is of great significance to research on zinc flotation working condition recognition and fault diagnosis. A blurry froth image includes not only the air–water fogs and dust produced in industrial sites, but also motion blur caused by camera vibrations. However, due to the redundancy and complexity of froth images, obtaining satisfactory results for motion-blurred bubble images by using existing methods is difficult. Therefore, we propose a deblurring method called filter-spatiotemporal-DeblurGAN (FS-DeblurGAN). First, the filter mechanism is used to screen the blurred froth image to solve the problem of high computing difficulty caused by large data volume. Second, the variable-order fractional differential operator is used to enhance the froth images to solve the problems of unclear edge and low contrast. Finally, we use the DeblurGAN method as a deblurring generator and effectively combine 3D and 2D streams to retrieve the spatiotemporal information of the froth images. Extensive experiments on several flotation froth datasets show that the proposed method achieves an excellent deblurring effect. The comparison experiment shows that the proposed method can better adapt to froth images under different conditions.

External IDs:doi:10.1140/epjs/s11734-022-00459-z