Targeting the issue of low efficiency and accuracy in manually detecting bubbles within the transparent layer of quartz crucibles,this paper introduces a deep fusion bubble detection algorithm(DFBDA).Firstly,an intelligent robotic arm is employed to automatically capture video images of bubbles within the transparent layer of quartz crucibles using an industrial microscope.Next,a bubble segmentation and contour extraction algorithm,employing fused morphological transformations,ensures accurate bubble segmentation.Lastly,employing a deep fusion technique,similar bubbles across multiple frames are merged, enabling a precise three-dimensional reconstruction of bubble positions. Experimentation across various crucible samples validates the method,demonstrating an average bubble identification accuracy of 98.7%,meeting both speed and precision requirements in industrial settings.