设施农业典型地物改进Faster R-CNN识别方法

王兴,康俊锋,刘学军,王美珍,张超

Improving the Faster R-CNN Method for Recognizing Typical Objects of Modern Agriculture based on Remote Sensing Imagery

WANG Xing,KANG Junfeng,LIU Xuejun,WANG Meizhen,ZHANG Chao

表2 基础特征提取网络结构

Tab. 2 Basic features extraction network structure

残差块	特征图尺寸宽度×长度,维度	残差内部结构宽度×长度,维度
残差块	特征图尺寸宽度×长度,维度	主径	捷径
残差单元 1	224×224, 64	$[\begin{array}{l} 1 \times 1, \\ 3 \times 3, \\ 1 \times 1, \end{array} \begin{array}{l} 16 \\ 16 \\ 32 \end{array}]$	224×224, 32
最大池化	224×224, 32
残差单元 2	112×112, 128	$[\begin{array}{l} 1 \times 1, \\ 3 \times 3, \\ 1 \times 1, \end{array} \begin{array}{l} 32 \\ 32 \\ 64 \end{array}]$	112×112, 64
最大池化	112×112, 64
残差单元 3	56×56, 256	$[\begin{array}{l} 1 \times 1, \\ 3 \times 3, \\ 1 \times 1, \end{array} \begin{array}{l} 64 \\ 64 \\ 128 \end{array}]$	56×56, 128
最大池化	56×56, 128
残差单元 4	28×28, 512	$[\begin{array}{l} 1 \times 1, \\ 3 \times 3, \\ 1 \times 1, \end{array} \begin{array}{l} 128 \\ 128 \\ 256 \end{array}]$	28×28, 256
最大池化	28×28, 256
残差单元 5	14×14, 1024	$[\begin{array}{l} 1 \times 1, \\ 3 \times 3, \\ 1 \times 1, \end{array} \begin{array}{l} 256 \\ 256 \\ 512 \end{array}]$	14×14, 512
最大池化	14×14, 512