The study of detecting method for impurity in transparent liquid

Abstract

How to detect and identify various kinds of impurities in transparent liquid effectively using machine vision is still a difficult task which has not been solved. The existing significant problems include the slow recognition speed, the lower recognition accuracy and the false and missing detection. Therefore, this paper presents a new impurity detection method. First, cluster the images captured using fuzzy c-means clustering algorithm and combine the clustering results with the improved least squares filtering to achieve impurity image background suppression. Then the method based on target counter area and prior knowledge is used to achieve target detection and tracking between frames. Finally, grading recognition strategy is used to improve the identification effect. In the pre-identifying stage, targets gray, size and position are used to achieve detection too-large impurities and normal small particles quickly. In the second stage, fuzzy least squares support vector machine is adopted to recognize the impurities which are similar with bubbles. The experimental results show the method proposed in this paper can achieve a 98.6% recognition rate while the false detecting rate is only 0.53% and the recognition time can be decreased to 100ms.