Overall training framework of the proposed method. The framework consists of two parts: supervised learning and pseudosupervised learning. In the supervised part, labeled data are used to train the student model, and a boundary loss is computed by performing edge detection on the ground truth. In the pseudosupervised part, unlabeled samples are input to generate pseudolabels through a network teacher (updated via the EMA student model) and a rule teacher (AOFlagger). High-confidence pseudolabels are then selected through an adaptive threshold mechanism and used to guide learning on unlabeled data.