Particle size detection and precise microstructure segmentation based on industrial CT are the core prerequisites for controlling pellets quality，elucidating the roasting mechanism，and ensuring the smooth operation of blast furnace.A systematic overview on the evolutionary process of pelletizing particle size detection technologies，from traditional manual screening and machine vision to intelligent detection methods based on deep learning，is proposed Computer vision application frameworks are focused on，while an in depth analysis of the algorithmic features and application effects of key technologies such as image enhancement，segmentation of agglomerated pelletizing，and accurate extraction of multiple phases from CT scan images is providedIn order to objectively evaluate the performance of the model，a unified pelletizing CT image dataset is constructed，and the multiphase segmentation effect of mainstream deep learning models is quantitatively compared. On this basis，the technical bottlenecks in complex industrial sites and microscopic imaging scenarios are deeply analyzed，including the difficulty of adhesion separation，strong interference between light and dust，blurred phase boundaries，and insufficient generalization ability of the model Furthermore，the future development trends such as customized depth models，multi modal fusion，macro and micro multi scale correlation and online closed loop control are prospected.The results show that it is difficult for traditional detection methods to take into account the accuracy of macroscopic particle size detection and the robustness of microscopic matter segmentation.The deep learning model based on the U Net series，attention mechanism and Transformer can significantly improve the segmentation accuracy of adherent pelletizing and the multiphase recognition effect of CT images of pelletizing，providing reliable support for the full dimensional quality evaluation of pelletizing.