A Novel Approach for Automatic Detection of Concrete Surface Voids Using Image Texture Analysis and History-Based Adaptive Differential Evolution Optimized Support Vector Machine

Table 4

Prediction result comparison.

Phase	Indices	L-SHADE-SVM-SVD		DCNN-Rmsprop		DCNN-Adam		DCNN-Sgdm		MB-BPNN
		Mean	Std.	Mean	Std.	Mean	Std.	Mean	Std.	Mean	Std.
Training	CAR (%)	97.417	0.222	87.922	2.585	89.211	1.937	87.756	6.165	86.763	6.889
	TP	437.200	4.938	369.000	25.984	390.250	17.693	375.650	40.359	333.900	63.570
	TN	439.550	5.753	422.300	20.683	412.650	21.755	414.150	21.866	360.200	9.540
	FP	9.550	1.605	27.700	20.683	37.350	21.755	35.850	21.866	66.100	63.570
	FN	13.700	1.302	81.000	25.984	59.750	17.693	74.350	40.359	39.800	9.540
	Precision	0.979	0.004	0.934	0.044	0.916	0.042	0.912	0.064	0.835	0.159
	Recall	0.970	0.003	0.820	0.058	0.867	0.039	0.835	0.090	0.895	0.012
	NPV	0.970	0.003	0.842	0.041	0.875	0.029	0.851	0.065	0.901	0.024
	F1 score	0.974	0.002	0.871	0.030	0.889	0.019	0.870	0.075	0.852	0.132
Testing	CAR (%)	92.600	2.761	88.350	3.133	86.900	4.204	86.800	6.178	85.700	7.248
	TP	46.200	4.873	41.700	3.246	42.200	3.792	40.900	4.767	40.900	8.130
	TN	46.400	5.305	46.650	2.323	44.700	3.326	45.900	2.674	44.800	2.587
	FP	4.500	1.878	3.350	2.323	5.300	3.326	4.100	2.674	9.100	8.130
	FN	2.900	2.125	8.300	3.246	7.800	3.792	9.100	4.767	5.200	2.587
	Precision	0.911	0.037	0.929	0.045	0.893	0.057	0.909	0.065	0.818	0.163
	Recall	0.942	0.040	0.834	0.065	0.844	0.076	0.818	0.095	0.890	0.042
	NPV	0.940	0.044	0.852	0.049	0.856	0.059	0.839	0.067	0.896	0.052
	F1 score	0.926	0.027	0.877	0.036	0.865	0.044	0.859	0.076	0.839	0.137