Abstract: Taking advantages of the SIFT(scale invariant feature transform) features,the problem that the bubble velocity feature is difficult to extract is solved after the SIFT feature matching is introduced into it.The extracted results in industrial flotation plant demonstrate that this method is insensitive to froth affine transformation and illumination variation,and is able to obtain the velocity features under various flotation states.After the observation of the froth surface visual features in a light metal mineral flotation plant within half a year,the relation between the froth velocity feature and flotation indexes is discussed,which lays the foundation for the machine vision based optimal control in flotation operation.