Triangular mesh model registration is an important part of industrial automation detection software. The registration accuracy has an important influence on mechanical parts' shape and position tolerance. Aiming to solve the problems of low accuracy and poor robustness of the automatic registration of triangular mesh models, we propose a segmentation method for enhanced features in the automatic registration of triangular mesh models for mechanical parts. First, the K value of the feature segmentation of the triangular mesh model was determined, and the Laplacian matrix determined the seed points for iterative initialization. Second, the appropriate region shape agent and cost function were used to accelerate the process and perform multi-source iterative clustering to obtain the intended feature segmentation results. Finally, based on the feature segmentation results of the triangular mesh model, the coarse registration based on the singular value decomposition method was performed, then the fine registration was performed according to the EM-ICP. The experimental results show that the proposed method reduces registration error by 25.2% and shortens the automatic registration time by 62.6%, compared with the traditional feature descriptor coarse registration combined with ICP fine registration method. This effectively improves the accuracy and efficiency of the automatic registration of the triangular mesh model.