Vision-Based Satellite Recognition and Pose Estimation Using Gaussian Process Regression

Table 2

Satellite recognition results on BUAA-SID 1.0 and comparison with [19, 31]. Results are in bold if the proposed method performs the best and in italic if the second best.

Number of training image	Representation	Method	Accuracy (%)
80	HU	SVM [31]	48.93
		Kernel method [19]	55.50
		Our GPR method	68.91
80	FD	SVM [31]	34.73
		Kernel method [19]	53.86
		Our GPR method	59.25
80	Elastic net sparse coding [31]	SVM [31]	82.07
	HOG	Kernel method [19]	97.13
	HOG	Our GPR method	84.98
90	HU	SVM [31]	50.50
		Kernel method [19]	56.04
		Our GPR method	69.67
90	FD	SVM [31]	34.36
		Kernel method [19]	54.25
		Our GPR method	59.52
90	Elastic net sparse coding [31]	SVM [31]	83.25
	HOG	Kernel method [19]	96.60
	HOG	Our GPR method	85.78
100	HU	SVM [31]	53.08
		Kernel method [19]	55.11
		Our GPR method	70.22
100	FD	SVM [31]	35.92
		Kernel method [19]	54.90
		Our GPR method	60.57
100	Elastic net sparse coding [31]	SVM [31]	87.54
	HOG	Kernel method [19]	97.65
	HOG	Our GPR method	85.40