利用单晶炉中特殊结构在硅熔液上的倒影图像目标, 通过图像处理进行高温硅熔体液面变化的非接触式检测。首先比较不同图像分割方法对倒影图像处理的效果, 选取最优图像分割方法处理不同单晶硅液位的倒影图像, 其次采用不同角度、不同尺寸ROI(感兴趣区域)对图像进行分割, 并提取分割后的目标像素面积, 最后利用像素面积变化与对应液位变化之间的比值系数和像素面积变化的标准差相结合的评价函数对图像处理效果进行对比。结果表明, 采用动态局部阈值分割法和大尺寸且垂直于相机视角的ROI对图像进行处理, 使得评价函数值最大, 兼顾了单晶硅液位检测的灵敏度和精密度。

The non-contact detection of the melt level change of the high-temperature silicon melt is performed by image processing using the reflection image target of the special structure on the silicon melt in the single crystal furnace. Firstly, the results of different image segmentation methods on the reflection image are compared. Secondly, the optimal image segmentation method to process the reflection images at different melt levels is applied. Then, different angles and different sizes of ROI (region of interest) to segment the image are used to extract the segmented target pixel area. Finally, the image segmentation results are compared by using an evaluation function combining the ratio coefficient between the pixel area change and the corresponding liquid level change and the standard deviation of the pixel area change. The results show that the value of the evaluation function is the largest, which takes into accounts the sensitivity and precision of melt level detection, when the dynamic local threshold segmentation method and the large-scale ROI perpendicular to the camera angle are used to process the image.