Table 4. ResNet50 analysis results between original images and rotated images

Model	Original Image	Top1	Top2	Top3	Top4	Top5
ResNet50 in 224 pixel
	X102	X95	X102_90	X95_90	X133_90	X40
		0.994	0.993	0.992	0.983	0.98
	X165	X163	X197	X274	X137	X130
		0.998(0.99813)	0.998(0.99809)	0.998(0.99801)	0.998(0.99800)	0.998(0.99795)
	X133	X133_270	X176_270	X176	X102_270	X89_270
		0.993	0.986	0.985	0.984	0.983
ResNet50 in 448 pixel
	X102	X102_90	X133	X133_90	X129_90	X129
		0.996(0.95929)	0.995(0.99498)	0.995(0.99486)	0.995(0.99467)	0.995(0.99460)
	X165	X165_180	X139	X224	X69	X139_180
		0.999(0.99948)	0.999(0.99933)	0.999(0.99929)	0.999(0.99922)	0.999(0.99904)
	X133	X133_270	X176_270	X102	X176	X102_270
		0.998	0.996	0.995(0.99498)	0.995(0.99478)	0.993

a: In imgaes “Xnumber_90”, 90 means that the image is rotated 90 degrees from the original
b: Correlation is a similarity value(the similarity rate is 100% as it approaches and it is rounded to the fourth decimal place