面向遥感影像场景分类的类中心知识蒸馏方法

刘潇, 刘智, 林雨准, 王淑香, 左溪冰

Class-centric Knowledge Distillation for RSI Scene Classification

LIU Xiao, LIU Zhi, LIN Yuzhun, WANG Shuxiang, ZUO Xibing

表2 师生网络结构及输出层特征信息

Tab. 2 Network structure of T/S models and Information about the features of the output layer

网络名称	ResNet50	ResNet18	MobileNetV2
卷积池化层	$7 \times 7,64, s t r i d e 2 + 3 \times 3 m a x p o o l, s t r i d e 2$	$3 \times 3,32, s t r i d e 2$
卷积层1	$[\begin{array}{l} 1 \times 1,64 \\ 3 \times 3,64 \\ 1 \times 1,256 \end{array}] \times 3$	$[\begin{array}{l} 3 \times 3,64 \\ 3 \times 3,64 \end{array}] \times 3$	$[\begin{array}{l} 1 \times 1,64 \\ 3 \times 3,64 \\ 1 \times 1,256 \end{array}] \times 3$
输出层特征1		$F \in R^{56 \times 56 \times 64}$	$F \in R^{56 \times 56 \times 256}$
卷积层2	$[\begin{array}{l} 1 \times 1,128 \\ 3 \times 3,128, s 2 \\ 1 \times 1,512 \end{array}] + [\begin{array}{l} 1 \times 1,128 \\ 3 \times 3,128 \\ 1 \times 1,512 \end{array}] \times 3$	$[\begin{array}{l} 3 \times 3,128, s 2 \\ 3 \times 3,128 \end{array}] + [\begin{array}{l} 3 \times 3,128 \\ 3 \times 3,128 \end{array}]$	$[\begin{array}{l} 1 \times 1,128 \\ 3 \times 3,128, s 2 \\ 1 \times 1,512 \end{array}] + [\begin{array}{l} 1 \times 1,128 \\ 3 \times 3,128 \\ 1 \times 1,512 \end{array}] \times 3$
输出层特征2	$F \in R^{28 \times 28 \times 512}$	$F \in R^{28 \times 28 \times 128}$	$F \in R^{28 \times 28 \times 512}$
卷积层3	$[\begin{array}{l} 1 \times 1,256 \\ 3 \times 3,256, s 2 \\ 1 \times 1,1024 \end{array}] + [\begin{array}{l} 1 \times 1,256 \\ 3 \times 3,256 \\ 1 \times 1,1024 \end{array}] \times 3$	$[\begin{array}{l} 3 \times 3,256, s 2 \\ 3 \times 3,256 \end{array}] + [\begin{array}{l} 3 \times 3,256 \\ 3 \times 3,256 \end{array}]$	$[\begin{array}{l} 1 \times 1,256 \\ 3 \times 3,256, s 2 \\ 1 \times 1,1024 \end{array}] + [\begin{array}{l} 1 \times 1,256 \\ 3 \times 3,256 \\ 1 \times 1,1024 \end{array}] \times 3$
输出层特征3	$F \in R^{14 \times 14 \times 1024}$	$F \in R^{14 \times 14 \times 256}$	$F \in R^{14 \times 14 \times 1024}$
卷积层4	$[\begin{array}{l} \begin{array}{l} 1 \times 1,, 512 \\ 3 \times 3,512, s 2 \end{array} \\ 1 \times 1,2048 \end{array}] + [\begin{array}{l} \begin{array}{l} 1 \times 1,, 512 \\ 3 \times 3,512 \end{array} \\ 1 \times 1,2048 \end{array}] \times 3$	$[\begin{array}{l} 3 \times 3,512, s 2 \\ 3 \times 3,512 \end{array}] + [\begin{array}{l} 3 \times 3,512 \\ 3 \times 3,512 \end{array}]$	$[\begin{array}{l} \begin{array}{l} 1 \times 1,, 512 \\ 3 \times 3,512, s 2 \end{array} \\ 1 \times 1,2048 \end{array}] + [\begin{array}{l} \begin{array}{l} 1 \times 1,, 512 \\ 3 \times 3,512 \end{array} \\ 1 \times 1,2048 \end{array}] \times 3$
输出层特征4	$F \in R^{7 \times 7 \times 2048}$	$F \in R^{7 \times 7 \times 512}$	$F \in R^{7 \times 7 \times 2048}$
全局池化层		$7 \times 7 a v e r a g e p o o l, f c, s o f t m a x$