Froth Image Enhancement And Segmentation Method And Its Application For Mineral Flotation

Froth flotation is a process for selectively separating ores, due to the differences in the physical and chemical properties of mineral sur-faces. Its froth surface characteristics are highly related to flotation condi-tions. Recently, China non-ferrous flotation mainly depends on operators by manipulating the operational parameters, based on observations of froth conditions. In the case of various feed ores, however, the manual observation has subjectivity and casualness; hence the optimal operation of flotation process would be significantly affected. Therefore, the ma-chine vision is introduced into the flotation process, so as to achieve the accurate measurement of froth parameters and quantitative description of production conditions, which is very important to the optimization of flo-tation process. The industrial environment is complicated with uneven il-lumination to flotation cells and much image noise. Moreover, the bub-bles are mixed together with irregular shapes, making it difficult to meas-ure the bubble size and shape. Therefore, to study extracting morpho-logical characteristics of froth images is very important to the industrial guidance and the efficiency in the utilization of mineral resources.Based on the mechanism analysis, the relationships between bubble sizes, bubble shapes and flotation conditions are studied. Combining the characteristics of froth images, a pre-processing method for froth images based on multi-scale analysis and watershed segmentation method for parameters measurement are introduced. Then, the proposed methods are successfully used in the foam image monitoring system of bauxite flota-tion process. The research and innovative achievements of the paper are reflected in the following aspects:(1) Considering the blurry images obtained from the flotation plant, caused by the non-uniform and non-directional froth motion, an image definition criterion of froth images based on adaptive window function is proposed. Further, it constructs a composite criterion function, which can effectively achieve a quantitative evaluation of froth images, obtain those good quality images. Experimental results show that the proposed method has good stability and practicality.(2) Since the obtained froth images are subject to uneven illumina-tion and shadow, a detailed analysis of Retinex theory is explained, then an adaptive multi-scale Retinex method combining the object features is proposed. The study adopts GCLM two-dimensional cross-entropy algo-rithm and fast binary algorithm respectively in image rough segmentation, which provide priori knowledge for the classification of froth area and adaptive selection of the structural elements. Through the identification and location of the strong contrast area, it avoids the edge halo phenom-ena caused by Retinex algorithm when dealing with strong contrast areas. This method effectively compensated for the uneven illumination of bub-ble image, and lays the foundation for analysis and processing of subse-quent images.(3) Considering the bubble image low-contrast, susceptible to envi-ronmental noise and light impact, multi-scale geometric analysis method of froth flotation image enhancement is introduced. Firstly, an improved multi-scale geometric analysis method is constructed to complete the foam image decomposition to achieve translational invariance of the sig-nal, avoiding the image edge blurring. Then, since the difference between decomposition coefficients for the low frequency sub-band is very small and severely affected by illumination, a multi-scale Retinex algorithm is proposed to improve its brightness uniformity. While for the high fre-quency subband, a nonlinear enhanced function is constructed based on the statistical model of decomposition coefficients. The method can im-prove the image brightness uniformity, enhance the weak edges, maintain strong edge information, eliminate the noise, and significantly improve the visual effects of the bubble images. It solves the problems of insuffi-cient segmentation caused by blurry edges and image noise.(4) Considering the mixed and adhesion characteristics of bubbles, the watershed segmentation method for parameter measurement is intro-duced. The fast threshold method is used to complete the coarse image segmentation, and then to identify and deal with transparent windows, black holes and narrow bright band by combining spatial relationships and LBPV texture characteristics, with the extraction of transparent win-dows and black holes index. Then an extraction method of multi-scale digital morphological markers based on adaptive region is proposed, fol-lowed by the multi-scale gradient image segmentation using the improved watershed method. The proposed method is oriented to object semantic features, combines the image segmentation, feature extraction and recog-nition together and improves the algorithm robustness. It also avoids the influences of working condition fluctuations and mixing, uneven bubbles on segmentation results, as well as reduces the insufficient and over seg-mentation areas.(5) This paper studies a bauxite flotation plant, and introduces a monitoring system of flotation process based on machine vision, which achieve the network sharing and transportation of data and operational information. Through the optimization of extraction function of charac-teristics parameters, it improves the robustness and accuracy of the real-time monitoring system. Furthermore, it greatly improved the pro-duction efficiency, laying the foundation for optimal control of flotation process.Figures(76), tables(6), references(157).