Measurement of Irregular Crystal Sizes Based on Improved MLESAC

To address the low accuracy in measuring crosssectional dimensions of irregular crystals grown via the Czochralski method, an improved MLESAC algorithm (DC-MLESAC) based on density clustering-guided sampling and adaptive likelihood iteration is proposed for ellipse fitting of crystal crosssections to measure major and minor axes. The method identifies core regions of feature points via DBSCAN clustering to construct initial models, avoiding outlier interference; dynamically optimizes likelihood parameters through iterative integration of residual median absolute deviation (MAD) and outlier ratio, overcoming the reliance on manually preset parameters in traditional methods; and introduces locally continuous weighted log-likelihood to enhance contributions of continuous edge points and suppress isolated noise. Experiments demonstrate that the proposed method achieves high accuracy and stability, with major and minor axis errors of 0.63 mm and 0.41 mm, respectively, exhibiting higher precision and robustness than other algorithms.