目前流行的深度学习库有Caffe，Keras，Theano，本文采用谷歌开源的曾用来制作AlphaGo的深度学习系统Tensorflow。
1：安装Tensorflow
最早TensorFlow只支持mac和Linux系统，目前也支持windows系统，但要求python3.5 (64bit)版本。TensorFlow有cpu和gpu版本，由于本文使用服务器是NVIDIA显卡，因此安装gpu版本，在cmd命令行键入
pip install --upgrade tensorflow-gpu
如果出现错误“Cannot remove entries from nonexistent file”，执行以下命令
“pip install --upgrade -I setuptools”，安装成功出现以下界面
2：安装CUDA库
用gpu来运行Tensorflow还需要配置CUDA和cuDnn库，
用以下link下载win10(64bit)版本CUDA安装包，大小约为1.2G https://developer.nvidia.com/cuda-downloads
安装cuda_8.0.61_win10.exe，完成后配置系统变量，在系统变量中的CUDA_PATH中，加上三个路径， C:Program FilesNVIDIA GPU Computing ToolkitCUDAv8.0
C:Program FilesNVIDIA GPU Computing ToolkitCUDAv8.0in
C:Program FilesNVIDIA GPU Computing ToolkitCUDAv8.0inlibx64
3：安装cuDnn库
用以下link下载cuDnn库
https://developer.nvidia.com/cudnn
下载解压后，为了在运行tensorflow的时候也能将这个库加载进去，我们要将解压后的文件拷到CUDA对应的文件夹下C:Program FilesNVIDIA GPU Computing ToolkitCUDAv8.0
4：测试安装
在Pycharm新建一个py文件键入
import tensorflow as tf
hello = tf.constant('Hello world, TensorFlow!')
sess = tf.Session()
print(sess.run(hello))
如果能够输出'Hello, TensorFlow!'像下面这样就代表配置成功了。
1: 导入必要的模块
import sys
import cv2
import numpy as np
import tensorflow as tf
2: 定义CNN的基本组件
按照LeNet5 的定义，采用32*32 图像输入，CNN的基本组件包括卷积C1，采样层S1、卷积C2，采样层S2、全结合层1、分类层2
3：训练CNN
将输入图像缩小至32*32大小，采用opencv中的resize函数
其变换参数有
CV_INTER_NN - 最近邻插值,
CV_INTER_LINEAR - 双线性插值 (缺省使用)
CV_INTER_AREA - 使用象素关系重采样。当图像缩小时候，该方法可以避免波纹出现。当图像放大时，类似于 CV_INTER_NN 方法..
CV_INTER_CUBIC - 立方插值.
output=cv2.resize(img,(32,32),interpolation=cv2.INTER_CUBIC)
核心代码如下：
class CNNetwork:
NUM_CLASSES = 2 #分两类
IMAGE_SIZE = 28
IMAGE_PIXELS = IMAGE_SIZE*IMAGE_SIZE*3
def inference(images_placeholder, keep_prob):
def weight_variable(shape):
initial = tf.truncated_normal(shape, stddev=0.1)
return tf.Variable(initial)
def bias_variable(shape):
initial = tf.constant(0.1, shape=shape)
return tf.Variable(initial)
def conv2d(x, W):
return tf.nn.conv2d(x, W, strides=[1, 1, 1, 1], padding='SAME')
def max_pool_2x2(x):
return tf.nn.max_pool(x, ksize=[1, 2, 2, 1],
strides=[1, 2, 2, 1], padding='SAME')
x_image = tf.reshape(images_placeholder, [-1, 28, 28, 1])
with tf.name_scope('conv1') as scope:
W_conv1 = weight_variable([5, 5, 3, 32])
b_conv1 = bias_variable([32])
h_conv1 = tf.nn.relu(conv2d(x_image, W_conv1) + b_conv1)
with tf.name_scope('pool1') as scope:
h_pool1 = max_pool_2x2(h_conv1)
with tf.name_scope('conv2') as scope:
W_conv2 = weight_variable([5, 5, 32, 64])
b_conv2 = bias_variable([64])
h_conv2 = tf.nn.relu(conv2d(h_pool1, W_conv2) + b_conv2)
with tf.name_scope('pool2') as scope:
h_pool2 = max_pool_2x2(h_conv2)
with tf.name_scope('fc1') as scope:
W_fc1 = weight_variable([7*7*64, 1024])
b_fc1 = bias_variable([1024])
h_pool2_flat = tf.reshape(h_pool2, [-1, 7*7*64])
h_fc1 = tf.nn.relu(tf.matmul(h_pool2_flat, W_fc1) + b_fc1)
h_fc1_drop = tf.nn.dropout(h_fc1, keep_prob)
with tf.name_scope('fc2') as scope:
W_fc2 = weight_variable([1024, NUM_CLASSES])
b_fc2 = bias_variable([NUM_CLASSES])
with tf.name_scope('softmax') as scope:
y_conv=tf.nn.softmax(tf.matmul(h_fc1_drop, W_fc2) + b_fc2)
return y_conv
if __name__ == '__main__':
test_image = []
filenames = []
for i in range(1, len(sys.argv)):
img = cv2.imread(sys.argv[i])
img = cv2.resize(img, (28, 28))
test_image.append(img.flatten().astype(np.float32)/255.0)
filenames.append(sys.argv[i])
test_image = np.asarray(test_image)
images_placeholder = tf.placeholder("float", shape=(None, IMAGE_PIXELS))
labels_placeholder = tf.placeholder("float", shape=(None, NUM_CLASSES))
keep_prob = tf.placeholder("float")
logits = inference(images_placeholder, keep_prob)
sess = tf.InteractiveSession()
saver = tf.train.Saver()
sess.run(tf.initialize_all_variables())
saver.restore(sess, "model.ckpt")
for i in range(len(test_image)):
pred = np.argmax(logits.eval(feed_dict={
images_placeholder: [test_image[i]],
keep_prob: 1.0 })[0])
pred2 = logits.eval(feed_dict={
images_placeholder: [test_image[i]],
keep_prob: 1.0 })[0]
print filenames[i],pred,"{0:10.8f}".format(pred2[0]),"{0:10.8f}".format(pred2[1])