Table 4.

Recognition effects of different CNN models on TUSP images.

Models		TPTI	LPTI	UTPLT	DTPLT	UTPRT	DTPRT	LPLT	LPRT	Macro average	Accuracy
ResNet-18	P	0.9815	0.9980	0.9410	0.9328	0.9512	0.9046	0.7852	0.8170	0.9139	0.9107
	R	0.9953	0.9970	0.9399	0.9241	0.9146	0.9408	0.7680	0.8272	0.9134
	F1	0.9883	0.9975	0.9401	0.9277	0.9322	0.9218	0.7753	0.8212	0.9130
ResNet-50	P	0.9845	0.9832	0.9026	0.8956	0.9179	0.8792	0.6884	0.7812	0.8791	0.8744
	R	0.9843	0.9890	0.9004	0.8842	0.8909	0.9100	0.7329	0.7348	0.8783
	F1	0.9843	0.9861	0.9006	0.8890	0.9030	0.8929	0.7078	0.7558	0.8775
ResNet-101	P	0.9922	0.9902	0.9117	0.9281	0.9040	0.8964	0.7492	0.7242	0.8870	0.8795
	R	0.9906	0.9910	0.9296	0.9041	0.9061	0.8876	0.6127	0.8261	0.8810
	F1	0.9913	0.9906	0.9203	0.9156	0.9048	0.8909	0.6710	0.7703	0.8818
VGG16	P	0.9937	0.9851	0.9328	0.8929	0.9039	0.8798	0.7190	0.7461	0.8817	0.8762
	R	0.9874	0.9900	0.8971	0.9221	0.8911	0.8813	0.6768	0.7815	0.8784
	F1	0.9905	0.9875	0.9132	0.9057	0.8956	0.8784	0.6963	0.7626	0.8787
ResNet-152	P	0.9938	0.9813	0.9247	0.9028	0.8826	0.8538	0.7800	0.7024	0.8777	08634
	R	0.9858	0.9910	0.9057	0.9258	0.8768	0.8569	0.5289	0.8370	0.8635
	F1	0.9897	0.9861	0.9143	0.9128	0.8773	0.8507	0.6042	0.7556	0.8613
InceptionV3	P	0.9907	0.9911	0.9359	0.9341	0.9220	0.8960	0.7430	0.7926	0.9007	0.8962
	R	0.9858	0.9920	0.9398	0.9313	0.9098	0.9140	0.7407	0.7870	0.9000
	F1	0.9881	0.9915	0.9374	0.9320	0.9150	0.9038	0.7398	0.7882	0.8995
MobileNet	P	0.9892	0.9921	0.9341	0.9294	0.9162	0.9118	0.7490	0.8028	0.9031	0.8986
	R	0.9937	0.9960	0.9347	0.9223	0.9199	0.8936	0.7613	0.7880	0.9012
	F1	0.9914	0.9940	0.9340	0.9254	0.9174	0.9018	0.7528	0.7932	0.9012
Xception	P	0.9844	0.9913	0.9298	0.9605	0.9504	0.9054	0.7634	0.7812	0.9083	0.9013
	R	0.9890	0.9900	0.9639	0.9061	0.9148	0.9406	0.7015	0.8315	0.9047
	F1	0.9867	0.9906	0.9461	0.9319	0.9318	0.9222	0.7258	0.8030	0.9047

The values in the table are the average of five-fold cross-validation.