立即学习:https://edu.csdn.net/course/play/26266/334242?utm_source=blogtoedu
LeNet模型核心概念:
1.卷积
input image-kernel-feature map
2 池化(下采样)尺寸减半所以下采样
max pooling & average pooling
3激活函数(ReLU)
LeNet逐层分析:
第一个卷积层C1
n(c):滤波器个数-No.of filters
(又称为卷积核)
F: Filter size
P: Padding
(是否需要填充,=0即不填充)
Stride: 步长
如何计算卷积后大小:
input image的宽度-kernel的宽度+1=feature map的宽度
32x32x1----28x28x6(F=5: 32-5+1)
Trainable parameters=Weight+Bias=156
weight = FxFx n x n+n
(? bias = n(c ) )
Connections =28x28x156(Trainable parameters)
第一个池化层S2
Trainable parameters=(coefficient +bias)xfilters (含有激活函数参数)
Connections =14x14x30