关键词: 浮选泡沫图像, 崩塌率检测, 非下采样 Shearlet 变换, 快速视网膜关键点, 形状全方向相似度

Abstract: A new froth collapse rate detection method by the fusion of fast retina keypoint (FREAK) and shape omnidirectional similarity in Nonsubsampled Shearlet transform (NSST) multiscale domain was proposed，consi-dering the difficulty to detect the froth collapse rate resulted from continuous flow and movement deformation． First-ly，two adjacent froth images were decomposed through NSST，froth bright spots are extracted by segmentation of low frequency subband image． And feature points were tested by direction modulus maxima detection and nonmaxi-mum suppression among multiscale high frequency subbands，then FREAK sampling pattern was improved and used for feature points description and matching． Secondly，potential collapsed bubbles were extracted according to the number of matching points that around the bright spots of previous frame，and then collapsed bubbles were selected from potential collapsed bubbles by using shape complexity feature and omnidirectional similarity detection
of bright spots between previous frame and next frame． Finally，the bubble collapse rate was calculated according to the detection results of collapsed bubble． Experimental results show that，the proposed method is affected little by nonuniform flow and movement deformation of bubbles and can effectively extract all collapsed bubbles． Be-sides，it achieves not only a better detection accuracy than that of existing methods，but also robustness of per-formance under different flotation working condition，thus this method meets the on-line detection need of flotation production．

Key words: flotation froth image, collapse rate detection, nonsubsampled Shearlet transform, fast retina key-point, shape omnidirectional similarity