<p>MnS inclusions in free-cutting steel were investigated using the BL16U2 beamline at the Shanghai Synchrotron Radiation Facility. Two-dimensional cross-sectional imaging of MnS inclusions was conducted through computed tomography. ImageJ software was employed to extract characteristic parameters from the MnS-containing two-dimensional images, followed by correlation analysis to identify the optimal combination of parameters for dataset construction. A multi-branch gated segmented integrated regression network (MBG-SIRN) was developed based on the PyTorch framework. This model adopts a stacked ensemble strategy, in which the multi-branch gated network and eXtreme gradient boosting (Xgboost) are used as base learners, with Xgboost also serving as the meta-learner. Furthermore, the Sparrow Search Algorithm was applied to automatically optimize hyperparameters such as learning rate and weight decay. The results demonstrate that the MBG-SIRN model exhibits outstanding predictive accuracy in estimating the number of MnS inclusions. Specifically, 98.3% of the samples achieved relative errors within the range of [0, 0.1], and the coefficient of determination approached 1, confirming the model’s ability to accurately quantify MnS inclusions.</p>