Dynamic Characteristic Extraction Method For Flotation Froth Image Based On SIFT

Mineral flotation is a complex physical and chemical process. Because of its complex and vulnerable craft mechanism, and many important parameters unable to be measured online, the flotation process is different to be described by a precise mathematical model. Now, the flotation operation in concentration plants is still based on the characteristics observed by the experienced operators from flotation froth. However, this method is greatly affected by subjective factor with low efficiency and large errors which makes the increase of instability and arbitrariness in the production, as a result of the decrease of mineral recovery. Therefore, it is very significant for the control of flotation to extract froth image's characteristics accurately, and build flotation control model to give the guidance for flotation process control.As for the mineral flotation process, a new dynamic characteristic extraction method of flotation froth image based on scale invariant feature transform(SIFT) is proposed in this study, considering the difficulty to detect the dynamic characteristic resulted from continuous movement of flotation froth image. Firstly, aiming at the problem of high rate of mismatch and low real-time performance in SIFT feature match of flotation froth images, a concept of motion matching region based on the unique characteristics of flotation froth is proposed, then according to the distribution range of magnitude and direction of bubble velocity to improve SIFT algorithm's matching conditions and using random sample consensus(RANSAC) to further eliminate the mismatching. Finally, according to the matching results of SIFT feature, the velocity characteristics and matching rate will be calculated, and a new method of extracting the collapse rate of the froth is proposed based on the matching result.To verify the validity of the proposed method in this paper, the flotation bubble real video images on a certain industrial scene are chosen to be tested. Firstly, the proposed approach is used for extracting and matching SIFT feature of froth video images, and compared with original methods to verify its superiority in matching points, matching accuracy and real-time. Then, the velocity, collapse and other characteristics will be extracted based on the accurate results of SIFT match, and the results will be analyzed to verify its environment adaptability under different working conditions. Comparing with former methods, the experimental results show the method proposed can accurately extract the dynamic characteristics of flotation froth, such as velocity, collapse rate, etc, and effectively eliminate the mismatching while reducing the computational complexity, improving the real-time even in complex industrial environments. At last, the flotation expert system using the method proposed in this paper illustrates a remarkable effect in the practical application.