The average performance and standard deviation of 30 independent runs (using different seed values) for LeNet, CNN-5, and CNN-8 are presented in Table 3 and the best performing method on each dataset is presented in boldface font. Apart from BrNoRo, LeNet has achieved the best performance on the other six datasets compared with CNN-5 and CNN-8.

Table 3:

Average accuracy (%) of three CNN methods on the seven texture images datasets ($x¯±s$).

BrNoRoBrWiRoOutexTC00OutexTC10KySinHwKyNoRoKyWiRo
LeNet 19.64 $±$ 6.56 $+$ 12.03$±$2.38$+$ 12.50$±$2.33$+$ 7.49$±$1.35$+$ 6.36$±$1.78$+$ 8.79$±$3.12$+$ 6.31$±$2.00$+$
CNN-5 21.36$±$6.56$+$ 12.01 $±$ 2.38 $+$ 5.03 $±$ 2.33 $+$ 4.81 $±$ 1.35 $+$ 6.09 $±$ 1.78 $+$ 5.39 $±$ 3.12 $+$ 4.80 $±$ 2.00 $+$
CNN-8 16.10 $±$ 3.97 $+$ 9.60 $±$ 2.64 $+$ 7.01 $±$ 3.39 $+$ 5.82 $±$ 1.78 $+$ 6.22 $±$ 1.95 $+$ 6.29 $±$ 2.39 $+$ 5.23 $±$ 1.81 $+$
BrNoRoBrWiRoOutexTC00OutexTC10KySinHwKyNoRoKyWiRo
LeNet 19.64 $±$ 6.56 $+$ 12.03$±$2.38$+$ 12.50$±$2.33$+$ 7.49$±$1.35$+$ 6.36$±$1.78$+$ 8.79$±$3.12$+$ 6.31$±$2.00$+$
CNN-5 21.36$±$6.56$+$ 12.01 $±$ 2.38 $+$ 5.03 $±$ 2.33 $+$ 4.81 $±$ 1.35 $+$ 6.09 $±$ 1.78 $+$ 5.39 $±$ 3.12 $+$ 4.80 $±$ 2.00 $+$
CNN-8 16.10 $±$ 3.97 $+$ 9.60 $±$ 2.64 $+$ 7.01 $±$ 3.39 $+$ 5.82 $±$ 1.78 $+$ 6.22 $±$ 1.95 $+$ 6.29 $±$ 2.39 $+$ 5.23 $±$ 1.81 $+$

Close Modal