基于张量特征的小样本图像快速分类方法

张艳莎, 冯夫健, 王杰, 潘凤, 谭棉, 张再军, 王林

Tensor feature⁃based faster classification network for few⁃shot learning

Yansha Zhang, Fujian Feng, Jie Wang, Feng Pan, Mian Tan, Zaijun Zhang, Lin Wang

表3 推理时间和分类准确率的对比

Table 3 Comparison of inference time and classification accuracy

Generate B tensor feature per class	Inference times (s)		Classification accuracy
Generate B tensor feature per class	1⁃shot	5⁃shot	1⁃shot	5⁃shot
$B = 1$	165 s (184 s)	188 s (204 s)	63.45±_0.77 (63.88±_0.85)	79.13±_0.56 (79.58±_0.59)
$B = 2$	166 s (184 s)	189 s (204 s)	63.72±_.79 (64.01±_0.83)	79.10±_0.56 (79.44±_0.61)
$B = 5$	166 s (189 s)	193 s (207 s)	63.79±_0.78 (64.12±_0.83)	79.11±_0.56 (79.49±_0.61)
$B = 10$	169 s (189 s)	195 s (209 s)	63.83±_0.79 (64.01±_0.85)	79.09±_0.56 (79.56±_0.63)
$B = 50$	172 s (194 s)	200 s (215 s)	63.31±_0.78 (63.67±_0.84)	78.88±_0.56 (79.18±_0.65)
$B = 100$	177 s (206 s)	204 s (227 s)	62.84±_0.78 (63.66±_0.86)	78.53±_0.55 (78.89±_0.65)
$B = 500$	244 s (304 s)	268 s (319 s)	63.84±_0.79 (64.11±_0.86)	78.74±_0.55 (79.38±_0.65)
$B = 1000$	334 s (442 s)	356 s (455 s)	64.34±_0.79 (64.69±_0.83)	78.70±_0.55 (79.48±_0.60)