Fig 1.
Non-support vectors become support vectors (H1 is the initial hyperplane. H3 is the final hyperplane. A1, A2, A3, A4, B1, B2, B3 and B4 are samples).
Fig 2.
The distance between the semi-labeled sample and the hyperplane is far but the classification result is still wrong (H1 is the initial hyperplane. H3 is the final hyperplane. A1, A2, A3, A4, B1, B2, B3 and B4 are samples).
Fig 3.
Flow-chart of “soft-start.”
Fig 4.
Flow-chart of incremental semi-supervised learning.
Fig 5.
Flow-chart of new labeled samples’ learning.
Fig 6.
Flow-chart of our classification algorithm based on incremental semi-supervised SVM.
Fig 7.
The accuracy rate of the testing set with different algorithms.
Fig 8.
All the semi-labeled are introduced correctly but the accuracy rate declines (H1 is the initial hyperplane, H3 is the final hyperplane and H2 is the hyperplane at a certain moment in the learning process.
A1, A2, A3, A4, B1, B2, B3 and B4 are samples. A* represents a testing sample in class A. Ag* and Bg* represent a group of testing samples in class A and class B, respectively).
Fig 9.
The accuracy rate of each new batch with different algorithms.