Figure 9
Semi-supervised learning process with the confirmed self-training and active learning (CST + AL) method. Two semi-supervised learnings, CST and AL, were performed in parallel using the same predicted labeled results. With the threshold and criteria adopted in this study, the same image was never selected as an additional candidate by both methods. All labeled images picked by both methods were added to the next set of teacher data. Then, they were subtracted from current unlabeled data to create the next set of unlabeled data. The rectangular frame shows automatic processing by a hand-made program written in Python.
