Table 2. Comparison of SVM and softmax classifier with FC7 high-level features extracted by the CNN.
Method | Classification | Area grading | Density grading | Location grading |
---|---|---|---|---|
CNN | 96.85% | 90.29% | 92.23% | 88.35% |
CNN_FC7 (Linear SVM)* | 96.85% | 90.29% | 93.20% | 89.32% |
CNN_FC7 (RBF SVM)§ | 96.85% | 90.29% | 92.23% | 90.29% |
log2c = -1; log2g = -8 | log2c = -6; log2g = -10 | log2c = -6; log2g = -8 | log2c = -5; log2g = -8 |
Footnotes: CNN (convolutional neural network); SVM (support vector machine); CNN_FC7 (the seventh fully-connected layer features of CNN); RBF (radial basis function);
*(linear kernel function SVM classifier);
§(RBF SVM classifier); c (cost parameter of RBF); g (gamma parameter of RBF).