Improving the adaptive filtration of the foam surface glare trend

Abstract

One of the important processes in the production of potash fertilizers is the froth flotation process. The quality of the final product depends significantly on the quality of the flotation. Technical vision is successfully used to control the flotation process. However, the existing methods of processing the video stream are inapplicable for controlling the process of flotation of potash ore due to the large scatter of statistical characteristics from one processed frame to another. This article discusses the use of nonblind filters to process streaming data. It is concluded that their application causes problems in identifying the moment of the beginning of the deviation. Based on this, the aim of the work is to reduce the noise level without affecting the identification of the transient, in other words, to improve the identification of the beginning of the transient by means of tunable blind filtering. It is proposed to recognize sets of N consecutive frames instead of single ones. For this, for each N frame, the number of bubbles, the average and median distances between them, and the average values of illumination and color components were calculated. From these calculations, it was concluded that the use of the arithmetic mean number of flares from N frames did not lead to an effective, significant reduction in the noise level. Therefore, it was proposed to use a different vector norm. As a result, an effective method for adaptive filtering of the trend of the number of highlights has been developed. On the materials of real video filming, a study was made of the change in noise from the number of frames. The results obtained show that the proposed method can reduce the standard deviation by 10-25% for different surveys. This proves the possibility of using the developed method for processing video streams both in laboratory and in industrial conditions.