A Flotation Froth Image Segmentation Based On An Improved Adaptive Weight FCM And Its Application

Mineral flotation is a vulnerable process to a variety of different physical and chemical factors, which are difficult to be separated; moreover, its craft mechanism is complex to be described by a precise mathematical model. Thus the flotation process is generally adjusted in real time through froth surface characteristics observed by the experienced operators. However, the process can’t run optimally and the mineral recovery decreases due to the uncertainty of flotation operation. Therefore it has a very practical significance to study the froth image conditions recognition, especially the extraction of size characteristics closely related to the flotation variables and performance indicators, and give the guidance for flotation process control.In this study, a new segmentation method based on an improved adaptive weight fuzzy C-means (AWFCM) is proposed considering the difficulty of the existing methods resulted from the diversity of noises and irregularity of bubble boundaries and particularity of gray level distribution of the froth images. Firstly, the image samples based on No-Reference Structural Sharpness (NRSS) are selected and image noise is removed by combining multi-resolution analysis and wavelet threshold denoising algorithm. Secondly, an adaptive weight FCM based on the particularity of gray level distribution is proposed for the image coarse division, and the morphological operations is used to denoise and smooth the image. Finally, the image is reconstructed to the gray-scale image by the distance transformation and the watershed algorithm is used to get waterline for every bubble. The bubble characteristics can be obtained from the segmented image, which can be used for condition recognition of the flotation process and giving the scientific guidance for the flotation control process.To verify the effectiveness and superiority of the proposed method, the flotation froth image and dosage data coming from a certain mineral plant are chosen to conduct the validation experiments. Firstly, the proposed method is used for flotation froth image to verify its validity. Then the proposed method in this paper is compared with the other three methods to verify its superiority. Finally, a conditions recognition model based on machine learning is established to find the relationship between the characteristics of flotation froth image and process variables to optimize flotation dosing control. The experimental results show that the proposed method has improved the segmentation accuracy significantly compared to the original FCM and other methods, which overcomes the over-segmentation and under-segmentation problem of the existing algorithms. As such, each bubble is extracted from the flotation image of irregular boundaries accurately that provides important information for the flotation control.