Abstract
Aiming at detecting and locating bubbles in transparent layer of quartz crucible during the growth of monocrystalline silicon, an improved YOLOv5s model algorithm is proposed in this paper to achieve efficient and accurate crucible bubble detection. Optimize the anchor box through the K-means algorithm to adapt to the bubble dataset before and after the use of the crucible. Utilize the FasterNet backbone network to extract contextual information, enhancing target detection effectiveness. Additionally, the ECA attention mechanism is added to enhance information exchange and improve the accuracy of small target detection. Experimental results show that the improved algorithm performs significantly in bubble detection. For bubble detection before and after use, mAP increased by 2.1 % and 3.29 %, respectively. As usage time increases, the size, shape, and distribution of bubbles in the transparent layer of the quartz crucible change significantly, directly affecting the crucible’s effectiveness. Regular monitoring and evaluation of crucible bubble changes are crucial for maintaining stability and safety in the production process. Further research could explore how these findings can optimize crucible design and usage, improving performance and longevity.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call