8 卷积神经网络——解决参数太多问题
8-1 卷积函数使用
手动生成一个55的矩阵来模拟图片，定义一个22的卷积核，来测试tf.nn.conv2d函数里的不同参数，验证其输出结果。
程序：
import tensorflow as tf
#1 定义输入变量
# [batch, in_height, in_width, in_channels] [训练时一个batch的图片数量, 图片高度, 图片宽度, 图像通道数]
input = tf.Variable(tf.constant(1.0, shape=[1, 5, 5, 1]))#定义一个5*5大小1个通道的矩阵
input2 = tf.Variable(tf.constant(1.0, shape=[1, 5, 5, 2]))
input3 = tf.Variable(tf.constant(1.0, shape=[1, 4, 4, 1]))
#2 定义卷积核变量
# [filter_height, filter_width, in_channels, out_channels] [卷积核的高度，卷积核的宽度，图像通道数，卷积核个数]
filter1 = tf.Variable(tf.constant([-1.0, 0, 0, -1], shape=[2, 2, 1, 1]))
filter2 = tf.Variable(tf.constant([-1.0, 0, 0, -1, -1.0, 0, 0, -1], shape=[2, 2, 1, 2]))#卷积核为2*2矩阵，1ch输入，2ch输出
filter3 = tf.Variable(tf.constant([-1.0, 0, 0, -1, -1.0, 0, 0, -1, -1.0, 0, 0, -1], shape=[2, 2, 1, 3]))
filter4 = tf.Variable(tf.constant([-1.0, 0, 0, -1,
-1.0, 0, 0, -1,
-1.0, 0, 0, -1,
-1.0, 0, 0, -1], shape=[2, 2, 2, 2]))
filter5 = tf.Variable(tf.constant([-1.0, 0, 0, -1, -1.0, 0, 0, -1], shape=[2, 2, 2, 1]))
#3 定义卷积操作
# padding的值为‘VALID’，表示边缘不填充, 当其为‘SAME’时，表示填充到卷积核可以到达图像边缘
op1 = tf.nn.conv2d(input, filter1, strides=[1, 2, 2, 1], padding='SAME') # 1个通道输入，生成1个feature ma
op2 = tf.nn.conv2d(input, filter2, strides=[1, 2, 2, 1], padding='SAME') # 1个通道输入，生成2个feature map
op3 = tf.nn.conv2d(input, filter3, strides=[1, 2, 2, 1], padding='SAME') # 1个通道输入，生成3个feature map
op4 = tf.nn.conv2d(input2, filter4, strides=[1, 2, 2, 1], padding='SAME') # 2个通道输入，生成2个feature
op5 = tf.nn.conv2d(input2, filter5, strides=[1, 2, 2, 1], padding='SAME') # 2个通道输入，生成一个feature map
vop1 = tf.nn.conv2d(input, filter1, strides=[1, 2, 2, 1], padding='VALID') # 5*5 对于pading不同而不同
op6 = tf.nn.conv2d(input3, filter1, strides=[1, 2, 2, 1], padding='SAME')
vop6 = tf.nn.conv2d(input3, filter1, strides=[1, 2, 2, 1], padding='VALID') # 4*4与pading无关
#4 运行卷积操作
init = tf.global_variables_initializer()
with tf.Session() as sess:
sess.run(init)
print("op1:\n", sess.run([op1, filter1])) # 1-1 后面补0
print("------------------")
print("op2:\n", sess.run([op2, filter2])) # 1-2多卷积核 按列取
print("op3:\n", sess.run([op3, filter3])) # 1-3
print("------------------")
print("op4:\n", sess.run([op4, filter4])) # 2-2 通道叠加
print("op5:\n", sess.run([op5, filter5])) # 2-1
print("------------------")
print("op1:\n", sess.run([op1, filter1])) # 1-1
print("vop1:\n", sess.run([vop1, filter1]))
print("op6:\n", sess.run([op6, filter1]))
print("vop6:\n", sess.run([vop6, filter1]))
结果：
op1:
[array([[[[-2.],
[-2.],
[-1.]],
[[-2.],
[-2.],
[-1.]],
[[-1.],
[-1.],
[-1.]]]], dtype=float32), array([[[[-1.]],
[[ 0.]]],
[[[ 0.]],
[[-1.]]]], dtype=float32)]
------------------
op2:
[array([[[[-2., -2.],
[-2., -2.],
[-2., 0.]],
[[-2., -2.],
[-2., -2.],
[-2., 0.]],
[[-1., -1.],
[-1., -1.],
[-1., 0.]]]], dtype=float32), array([[[[-1., 0.]],
[[ 0., -1.]]],
[[[-1., 0.]],
[[ 0., -1.]]]], dtype=float32)]
op3:
[array([[[[-2., -2., -2.],
[-2., -2., -2.],
[-1., -1., -1.]],
[[-2., -2., -2.],
[-2., -2., -2.],
[-1., -1., -1.]],
[[-2., -1., 0.],
[-2., -1., 0.],
[-1., 0., 0.]]]], dtype=float32), array([[[[-1., 0., 0.]],
[[-1., -1., 0.]]],
[[[ 0., -1., -1.]],
[[ 0., 0., -1.]]]], dtype=float32)]
------------------
op4:
[array([[[[-4., -4.],
[-4., -4.],
[-2., -2.]],
[[-4., -4.],
[-4., -4.],
[-2., -2.]],
[[-2., -2.],
[-2., -2.],
[-1., -1.]]]], dtype=float32), array([[[[-1., 0.],
[ 0., -1.]],
[[-1., 0.],
[ 0., -1.]]],
[[[-1., 0.],
[ 0., -1.]],
[[-1., 0.],
[ 0., -1.]]]], dtype=float32)]
op5:
[array([[[[-4.],
[-4.],
[-2.]],
[[-4.],
[-4.],
[-2.]],
[[-2.],
[-2.],
[-1.]]]], dtype=float32), array([[[[-1.],
[ 0.]],
[[ 0.],
[-1.]]],
[[[-1.],
[ 0.]],
[[ 0.],
[-1.]]]], dtype=float32)]
------------------
op1:
[array([[[[-2.],
[-2.],
[-1.]],
[[-2.],
[-2.],
[-1.]],
[[-1.],
[-1.],
[-1.]]]], dtype=float32), array([[[[-1.]],
[[ 0.]]],
[[[ 0.]],
[[-1.]]]], dtype=float32)]
vop1:
[array([[[[-2.],
[-2.]],
[[-2.],
[-2.]]]], dtype=float32), array([[[[-1.]],
[[ 0.]]],
[[[ 0.]],
[[-1.]]]], dtype=float32)]
op6:
[array([[[[-2.],
[-2.]],
[[-2.],
[-2.]]]], dtype=float32), array([[[[-1.]],
[[ 0.]]],
[[[ 0.]],
[[-1.]]]], dtype=float32)]
vop6:
[array([[[[-2.],
[-2.]],
[[-2.],
[-2.]]]], dtype=float32), array([[[[-1.]],
[[ 0.]]],
[[[ 0.]],
[[-1.]]]], dtype=float32)]
8-2 sobel
通过卷积操作实现sobel算子，讲彩色图片生成带有边缘化信息的图片
程序：
import matplotlib.pyplot as plt # plt 用于显示图片
import matplotlib.image as mpimg # mpimg 用于读取图片
import numpy as np
import tensorflow as tf
#1 载入图片并显示
myimg = mpimg.imread('img.jpg') # 读取和代码处于同一目录下的图片
plt.imshow(myimg) # 显示图片
plt.axis('off') # 不显示坐标轴
plt.show()
print(myimg.shape)
print('-----------------------------------------------------------')
#2 定义占位符、卷积核、卷积op
full = np.reshape(myimg, [1, 3264, 2448, 3])
inputfull = tf.Variable(tf.constant(1.0, shape=[1, 3264, 2448, 3]))
filter = tf.Variable(tf.constant([[-1.0, -1.0, -1.0], [0, 0, 0], [1.0, 1.0, 1.0],
[-2.0, -2.0, -2.0], [0, 0, 0], [2.0, 2.0, 2.0],
[-1.0, -1.0, -1.0], [0, 0, 0], [1.0, 1.0, 1.0]], shape=[3, 3, 3, 1]))
op = tf.nn.conv2d(inputfull, filter, strides=[1, 1, 1, 1], padding='SAME') # 3个通道输入，生成1个feature ma
o = tf.cast(((op - tf.reduce_min(op)) / (tf.reduce_max(op) - tf.reduce_min(op))) * 255, tf.uint8)
print('-----------------------------------------------------------')
#3 运行卷积操作并显示
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
t, f = sess.run([o, filter], feed_dict={inputfull: full})
# print(f)
t = np.reshape(t, [3264, 2448])
plt.imshow(t, cmap='Greys_r') # 显示图片
plt.axis('off') # 不显示坐标轴
plt.show()
结果：
(3264, 2448, 3)
-----------------------------------------------------------
-----------------------------------------------------------
Process finished with exit code 0
8-3 池化函数使用
定义1个输入变量来模拟输入图片，4*4大小的2通道矩阵，并将其赋予指定的值。2个通道分别为4个0到4个3组成的矩阵，4个4到4个7组成的矩阵。
程序：
import tensorflow as tf
#1 定义输入变量
img = tf.constant([
[[0.0, 4.0], [0.0, 4.0], [0.0, 4.0], [0.0, 4.0]],
[[1.0, 5.0], [1.0, 5.0], [1.0, 5.0], [1.0, 5.0]],
[[2.0, 6.0], [2.0, 6.0], [2.0, 6.0], [2.0, 6.0]],
[[3.0, 7.0], [3.0, 7.0], [3.0, 7.0], [3.0, 7.0]]
])
img = tf.reshape(img, [1, 4, 4, 2])
#2 定义池化操作
pooling = tf.nn.max_pool(img, [1, 2, 2, 1], [1, 2, 2, 1], padding='VALID')
pooling1 = tf.nn.max_pool(img, [1, 2, 2, 1], [1, 1, 1, 1], padding='VALID')
pooling2 = tf.nn.avg_pool(img, [1, 4, 4, 1], [1, 1, 1, 1], padding='SAME')
pooling3 = tf.nn.avg_pool(img, [1, 4, 4, 1], [1, 4, 4, 1], padding='SAME')
nt_hpool2_flat = tf.reshape(tf.transpose(img), [-1, 16])
pooling4 = tf.reduce_mean(nt_hpool2_flat, 1) # 1对行求均值（1表示轴是列） 0 对列求均值
#3 运行池化操作
with tf.Session() as sess:
print("image:")
image = sess.run(img)
print(image)
result = sess.run(pooling)
print("reslut:\n", result)
result = sess.run(pooling1)
print("reslut1:\n", result)
result = sess.run(pooling2)
print("reslut2:\n", result)
result = sess.run(pooling3)
print("reslut3:\n", result)
flat, result = sess.run([nt_hpool2_flat, pooling4])
print("reslut4:\n", result)
print("flat:\n", flat)
结果：
image:
[[[[0. 4.]
[0. 4.]
[0. 4.]
[0. 4.]]
[[1. 5.]
[1. 5.]
[1. 5.]
[1. 5.]]
[[2. 6.]
[2. 6.]
[2. 6.]
[2. 6.]]
[[3. 7.]
[3. 7.]
[3. 7.]
[3. 7.]]]]
reslut:
[[[[1. 5.]
[1. 5.]]
[[3. 7.]
[3. 7.]]]]
reslut1:
[[[[1. 5.]
[1. 5.]
[1. 5.]]
[[2. 6.]
[2. 6.]
[2. 6.]]
[[3. 7.]
[3. 7.]
[3. 7.]]]]
reslut2:
[[[[1. 5. ]
[1. 5. ]
[1. 5. ]
[1. 5. ]]
[[1.5 5.5]
[1.5 5.5]
[1.5 5.5]
[1.5 5.5]]
[[2. 6. ]
[2. 6. ]
[2. 6. ]
[2. 6. ]]
[[2.5 6.5]
[2.5 6.5]
[2.5 6.5]
[2.5 6.5]]]]
reslut3:
[[[[1.5 5.5]]]]
reslut4:
[1.5 5.5]
flat:
[[0. 1. 2. 3. 0. 1. 2. 3. 0. 1. 2. 3. 0. 1. 2. 3.]
[4. 5. 6. 7. 4. 5. 6. 7. 4. 5. 6. 7. 4. 5. 6. 7.]]