tensorflow cnn 特征提取(代码)

import tensorflowas tf

import numpyas np

####数据集准备

batch_size =100; w =10; h =50; c =1

X = np.random.uniform(0,1,(batch_size, w, h, c))

Y = np.random.randint(0,2,(batch_size))

# print(X)

# print(Y)

####图模型--占位符

G_X = tf.placeholder(tf.float32, [None, w, h, c])

G_Y = tf.placeholder(tf.int32, [None])

##权重矩阵初始化

def weight_variable(shape):

##用正太分布来初始化权值

    initial = tf.truncated_normal(shape, stddev=0.1)

return tf.Variable(initial)

##偏量初始化

def bias_variable(shape):

##设置一个很小的偏量

    initial = tf.constant(0.1, shape=shape)

return tf.Variable(initial)

##卷积层定义

def con2d(x, W):

##默认strides[0]=strides[3]=1, strides[1]为x方向步长,strides[2]为y方向步长

    return tf.nn.conv2d(x, W, strides=[1,1,1,1], padding='SAME')

##池化层定义

def max_pool_2d(x):

return tf.nn.max_pool(x, ksize=[1,2,2,1], strides=[1,2,2,1], padding='SAME')

##cnn--卷积层 过滤器形状5x2x1设置8个

W_conv1 = weight_variable([5,2,1,8])

b_conv1 = bias_variable([8])

h_conv1 = tf.nn.relu(con2d(G_X, W_conv1) + b_conv1)

##cnn--池化层

h_pool1 = max_pool_2d(h_conv1)

###flatten层  扁平化处理

f_flatten = tf.reshape(h_conv1, [-1,5*25*8])

###全连接,处理成类似标签的样子

W_f1 = weight_variable([5*25*8, 1])

b_f1 = bias_variable([1])

h_res = tf.nn.sigmoid(tf.matmul(f_flatten, W_f1) + b_f1)

res = tf.reshape(h_res, [-1])

with tf.Session()as sees:

sees.run(tf.global_variables_initializer())

print(sees.run(h_conv1, feed_dict={G_X:X, G_Y:Y}))

print(h_conv1)

print(type(h_conv1))

print(sees.run(h_pool1, feed_dict={G_X:X, G_Y:Y}))

print(h_pool1)

print(type(h_pool1))

print(sees.run(res, feed_dict={G_X: X, G_Y: Y}))

print(res)

print(type(res))

print(G_Y)