Abstract: In the process of industrial production of sapphire crystals, the seeding step plays a vital role. Seeding needs to be performed under conditions where the temperature gradient is small and the temperature distribution tends to be stable. At present, the industrial production of sapphire mainly relies on manual experience to control the seed rod to achieve the crystal pilot operation, but the low accuracy of manual pilot operation will lead to poor quality of finished products and waste of resources. Therefore, this paper proposes a method based on sapphire visual spokes pattern recognition method to detect the free liquid surface state of sapphire melt, so as to realize an efficient crystal induction mechanism. In this method, classic skeletoning algorithm was used to refine the spoke pattern, and Harris operator was used to extract the characteristic information. The extracted characteristic information was put into the motion trajectory model to judge the stability of the melt, and the stability of the temperature distribution of the liquid surface was analyzed to achieve crystal induction. Experimental results show that this algorithm is effective, the efficiency of sapphire crystal seeding is greatly improved, and the yield of the finished product is also improved, which can effectively guide the industrial production of sapphire.

Key words: sapphire, seeding, temperature gradient, spoke pattern, pattern recognition, motion trace, skeletonization algorithm, feature information