Pellet particle size is one of the important criteria to measure pellet quality, and the identification of pellet particle size is an important basis for improving the pellet qualification rate and regulating the pelletizing parameters. In order to solve the problem that the pellet particle size cannot be accurately fed back and rely on manual experience in the process of pelletizing, a machine vision particle size recognition algorithm based on SAM is proposed. In this method, OTSU is used to automatically generate cue points, and the secondary optimization is done according to the cue point optimization mobile strategy. The cue points are input into the segmentation mask of each pellet predicted by SAM, and the contour of a single pellet is extracted after filtering and repairing the prediction mask. According to the loss function discriminant segmentation, the integrity of the pellet is obtained, and the particle size of the pellet is obtained by adaptively selecting the least squares method or the minimum circumscribed circle to fit the pellet profile. The results show that the recognition rate in the sparse distribution and adhesion distribution of finished pellet reaches 100% , and the recognition rate in the overlapping distribution can reach more than 96% . In the process of pelletizing, the recognition rate of fresh balls can reach 80%. Compared with the existing algorithm models, the proposed model has stronger stability and accuracy in pellet particle size recognition, and has a wider range of application scenarios. Therefore, the SAM-based pellet image particle size recognition algorithm can accurately identify the pellet particle size information, and the model has good generalization performance, which provides an effective means to improve the intelligence level of pellet process.