Robust Grape Detector Based on SVMs and HOG Features

Table 5

Evaluation of the detectors trained on T-3, set-up according to Table 3, on original datasets of environment type (E), where and . Following versions of the detectors evaluated using the measures (1a), (1b), and (1c): is detector without any conversion, is detector with the conversion according to ITU-R recommendation BT.601, and R is detector with the generalized conversion (2).

Kernel	Version	E-1			E-2			E-3			E-4			E-5
		Accuracy	Precision	Recall	Accuracy	Precision	Recall	Accuracy	Precision	Recall	Accuracy	Precision	Recall	Accuracy	Precision	Recall
Linear		86.50%	1.0000	0.7300	87.25%	0.9869	0.7550	85.50%	0.9671	0.7350	86.25%	0.9801	0.7400	84.50%	0.9859	0.7000
		88.50%	0.9873	0.7800	88.50%	0.9753	0.7900	87.50%	0.9573	0.7850	87.75%	0.9748	0.7750	87.00%	0.9868	0.7500
		88.25%	0.9693	0.7900	90.25%	0.9763	0.8250	88.75%	0.9586	0.8100	88.25%	0.9693	0.7900	87.75%	0.9809	0.7700
RBF		88.00%	0.9935	0.7650	89.50%	0.9938	0.7950	88.00%	0.9872	0.7700	87.75%	0.9935	0.7600	86.25%	1.0000	0.7250
		89.50%	1.0000	0.7900	90.25%	0.9879	0.8150	89.00%	0.9815	0.7950	89.25%	0.9937	0.7900	87.50%	1.0000	0.7500
		88.50%	0.9936	0.7750	89.50%	0.9877	0.8000	88.50%	0.9812	0.7850	88.00%	0.9634	0.7900	86.75%	0.9933	0.7400