神经网络之LeNet网络结构（Python+TensorFlow）

# -*- coding:utf-8 -*-
#
# LeNet model

import tensorflow as tf
import numpy as np


def get_weight(shape, name):
    return tf.Variable(tf.truncated_normal(shape, stddev=0.1), name=name)

def get_bias(shape, name):
    return tf.Variable(tf.constant(0.0, shape=shape), name=name)

def conv_layer(x, ks, out_units, name):
    with tf.variable_scope(name):
        in_units = x.get_shape().as_list()[-1]
        filt = get_weight([ks,ks,in_units,out_units], name='weight')
        bias = get_bias([out_units], name='bias')
        out  = tf.nn.relu(tf.nn.bias_add(tf.nn.conv2d(x, filt, [1,1,1,1], padding='VALID'), bias))
    return out

def fc_layer(x, out_units, name):
    with tf.variable_scope(name):
        in_units = np.prod(x.get_shape().as_list()[1:])
        x_flat = tf.reshape(x, [-1, in_units])
        weight = get_weight([in_units,out_units], name='weight')
        biases = get_bias([out_units], name='bias')
        out    = tf.nn.bias_add(tf.matmul(x_flat, weight), biases)
    return out


def lenet(images, num_class=10, keep_prob=0.5):
    # conv1
    conv1 = conv_layer(images, ks=5, out_units=32, name='C1')
    # pool2
    pool2 = tf.nn.max_pool(conv1, ksize=[1,2,2,1], strides=[1,2,2,1], padding='SAME', name='S2')
    # conv3
    conv3 = conv_layer(pool2, ks=5, out_units=64, name='C3')
    # pool4
    pool4 = tf.nn.max_pool(conv3, ksize=[1,2,2,1], strides=[1,2,2,1], padding='SAME', name='S4')
    # conv5
    conv5 = conv_layer(pool4, ks=5, out_units=120, name='C5')
    # fc6
    fc6 = fc_layer(conv5, out_units=84, name='F6')
    fc6_drop = tf.nn.dropout(fc6, keep_prob)
    # fc7
    out = fc_layer(fc6_drop, out_units=num_class, name='F7')

    return out