Toward Recognition of Easily Confused TCM Herbs on the Smartphone Using Hierarchical Clustering Convolutional Neural Network

Table 7

The classification accuracy of CNN and the hierarchical CNN method for each herb.

	CNN (%)	HCNN by AP algorithm (average) (%)	HCNN based on manual clustering (%)
A1	100.0	100.0	100.0
A2	98.0	100.0	100.0
B1	81.0	90.8	92.0
B2	93.0	94.9	95.0
C1	90.0	93.2	94.5
C2	91.5	93.0	93.0
D1	98.0	99.0	99.0
D2	98.0	99.5	99.5
E1	96.5	98.5	98.5
E2	94.5	98.0	98.0
F1	100.0	100.0	100.0
F2	98.0	99.0	99.0
G1	98.0	98.9	99.0
G2	93.0	97.3	97.5
H1	93.5	93.5	94.5
H2	88.5	94.8	97.0
I1	94.5	98.9	99.5
I2	97.5	98.0	98.0
J1	99.5	99.6	99.5
J2	100.0	99.7	100.0
K1	99.0	99.7	99.5
K2	99.0	99.6	99.5
L1	94.5	95.3	96.0
L2	100.0	100.0	100.0
Avg.	95.65%	97.54	97.85

A significant improvement for some herbs such as B1 and H2 in addition to a general improvement of average accuracy (about 2%) when the HCNN method is employed.