Table 6.

Results of comparative performance analysis of our proposed method with various baseline methods for cross data sets.

Data sets and our methods		F1	APa	ARb	ACCc	AUCd
MCe train and SZf test
	LBPg and SVMh,i [46]	0.496	0.492	0.5	0.492	0.69
	HoGj and SVMi [47]	0.664	0.695	0.635	0.639	0.762
	ShuffleNeti [43]	0.661	0.715	0.615	0.61	0.709
	InceptionV3i [44]	0.708	0.717	0.7	0.698	0.761
	MobileNetV2i [45]	0.613	0.678	0.559	0.565	0.78
	ResNet50i [29]	0.686	0.707	0.667	0.663	0.77
	ResNet101i [29]	0.674	0.677	0.671	0.672	0.772
	GoogLeNeti [20,21]	0.592	0.595	0.589	0.591	0.65
	Santosh and Antani [16]	—k	—	0.76	0.76	0.82
	Proposed	0.795	0.798	0.793	0.792	0.853
SZ train and MC test
	LBP and SVMi [46]	0.537	0.58	0.5	0.58	0.552
	HoG and SVMi [47]	0.559	0.573	0.546	0.594	0.601
	ShuffleNeti [43]	0.633	0.643	0.624	0.652	0.683
	InceptionV3i [44]	0.681	0.722	0.644	0.688	0.748
	MobileNetV2i [45]	0.668	0.772	0.589	0.652	0.797
	ResNet50i [29]	0.64	0.642	0.638	0.616	0.787
	ResNet101i [29]	0.641	0.726	0.574	0.638	0.698
	GoogLeNeti [20,21]	0.648	0.691	0.609	0.659	0.754
	Santosh and Antani [16]	—	—	0.79	0.78	0.85
	Proposed	0.811	0.808	0.813	0.797	0.873

^aAP: average precision.

^bAR: average recall.

^cACC: accuracy.

^dAUC: area under the curve.

^eMC: Montgomery County.

^fSZ: Shenzhen.

^gLBP: local binary pattern.

^hSVM: support vector machine.

ⁱWe also evaluated the performance of these models (for the cross data set) using our selected data sets and experimental protocol.

^jHoG: histogram of oriented gradients.

^k—: not available. The results were not provided in this comparative study for these performance metrics.