A Research Of The Classfiction Detection Technology Of Pearls Based On Machine Vision

At present, most of the pearl-producing enterprises in our countryadopt the manual way to roughly separate and select pearls. Because pearlsare small in size and large in amount, the classification of pearls is atroublesome and inefficient job. After the national classification standard ofcultured pearls is brought into force, it is a pressing demand forpearl-producing enterprises to classify pearls accurately and speedily. Bythe tender requests of a pearl-producing enterprise in Zhuji, this paper doesresearch on the application of machine vision to the classification of pearls.The main points are as follows:1. Combining the analysis of pearls' physics and chemical propertyand their optical property with the requests in the national classificationstandard, this paper makes an overall design of the classification system ofpearls. The design of image-capturing module is paid special attention to: a)The method of lighting from a high angle and in a single direction is used to balance the nonuniform reflection and illumination on a pearl's surface;b) Pearls are light and irregular in shape—thus a rotary device is designedwhich can ensure the rotation of pearls so that the entire surface can bedetected.2. This paper works out some relevant image processing algorithmsaccording to the requests of the classification of pearls: a) A weightedvector median filtering algorithm is proposed, which greatly reduces theoperation amount owing to the merits of vector median filtering; b) As thesolution to the nonuniform illumination on a pearl's surface, local thresholdsegmentation and connected area label are adopted to get a stable andaccurate segmentation; c) Colors are classified according to localplacement of the hue histogram, which leads to results more stable andaccurate than the HSI space; d) The shape and size of pearls are measuredby a fast algorithm based of the shape under a high precision condition; e)Surface perfection is detected by boundary description and colorsegmentation, and flaws are classified according to flaw distributionfeatures.This paper makes some exploration and attempts in the application ofmachine vision to the classification detection of pearls, and gains someachievements which build the foundation for the future commercialclassification system. The experiment results proved that machine visioncan achieve a relatively high accuracy and speediness in size and color detection, and that to a certain extent it can meet the requests in surfaceperfection detection.