CNN用于识别验证码程序

import tensorflow as tf
import numpy as np
import os
from PIL import Image
import matplotlib.pyplot  as plt
os.environ["TF_CPP_MIN_LOG_LEVEL"] = "3"

#设置随机种子
tf.set_random_seed(777)
#设置路径
train_path = r"D:\vcode_data\train"
test_path = r"D:\vcode_data\test"

#图片转为灰度
def convert2grey(img):
    if len(img.shape)>2:
        r,g,b, = img[:,:,0],img[:,:,1],img[:,:,2]
        grey = (r+g+b)/3
        return grey
    else:
        return img

#定义读取图片数据方法
def get_file(path,num):
    x_data = []
    y_data = []
    count = 0
    for item in os.listdir(path):
        #将读取的图片路径放入x_path 中
        img_path = path+"\\"+item
        img= plt.imread(img_path)
        x_data.append(convert2grey(img).reshape(60,160,1))
        #生成标签数据
        y_one_hot = np.eye(10)[list(map(int,item[:4]))].flatten()
        y_data.append(y_one_hot)
        # 每读取一张图片，计数+1
        count += 1
        #当读取图片数量达到num，就停止
        if count == num:
            break
    return np.array(x_data),np.array(y_data)

#分别获取训练集和测试集
train_x,train_y = get_file(train_path,100) #1000张作为训练集
test_x,test_y = get_file(test_path,100)     #100张作为测试集

#获取图片维度
height = train_x.shape[1]
width = train_x.shape[2]

#定义next_batch 方法
g_b = 0
def next_batch(size):
    global g_b
    x = train_x[g_b:g_b+size]
    y = train_y[g_b:g_b+size]
    g_b += size
    return x,y

#定义占位符
x = tf.placeholder(tf.float32,shape=[None,height,width,1])
y = tf.placeholder(tf.float32,shape=[None,40])

#卷积层1
w1 = tf.Variable(tf.random_normal([3,3,1,32]))
L1 = tf.nn.conv2d(x,w1,strides=[1,1,1,1],padding="SAME")
L1 = tf.nn.relu(L1)
L1 = tf.nn.max_pool(L1,ksize=[1,2,2,1],strides=[1,2,2,1],padding="SAME")

#卷积层2
w2 = tf.Variable(tf.random_normal([3,3,32,64]))
L2 = tf.nn.conv2d(L1,w2,strides=[1,1,1,1],padding="SAME")
L2 = tf.nn.relu(L2)
L2 = tf.nn.max_pool(L2,ksize=[1,2,2,1],strides=[1,2,2,1],padding="SAME")

#全连接层
dim = L2.get_shape()[1].value*L2.get_shape()[2].value*L2.get_shape()[3].value
L2_flat = tf.reshape(L2,shape=[-1,dim])

w3 = tf.Variable(tf.random_normal([dim,40]))
b = tf.Variable(tf.random_normal([40]))
logits = tf.matmul(L2_flat,w3)+b

#定义代价
loss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=logits,labels=y))
#优化器
optimizer = tf.train.AdamOptimizer(learning_rate=0.001).minimize(loss)
#定义精度
y_pred = tf.argmax(tf.reshape(logits,shape=[-1,4,10]),axis=2)
y_true = tf.argmax(tf.reshape(y,shape=[-1,4,10]),axis=2)
correct = tf.equal(y_pred,y_true)
accuracy = tf.reduce_mean(tf.cast(correct,tf.float32))

#定义会话
sess = tf.Session()
sess.run(tf.global_variables_initializer())

#开始迭代
training_epoch = 10
batch_size = 100
for epoch in range(training_epoch):
    avg_cost = 0
    total_batch = int(1000/batch_size)
    for i in range(total_batch):
        batch_x,batch_y = next_batch(batch_size)
        c,_ = sess.run([loss,optimizer],feed_dict={x:train_x,y:train_y})
        avg_cost += c/total_batch
    #精度
    acc = sess.run(accuracy,feed_dict={x:train_x,y:train_y})
    print("epoch:",epoch+1,"acc:",acc)

#打印测试集精度
print("测试集精度是：",sess.run(accuracy,feed_dict={x:test_x,y:test_y}))

#随机测试一个样本
np.random.seed(100)
r = np.random.randint(0,len(test_x)-1)
print("实际值是：",sess.run(tf.argmax(test_y[r].reshape(4,10),1)))
print("预测值是：",sess.run(tf.argmax(tf.reshape(logits,shape=[4,10]),1),feed_dict={x:test_x[r:r+1]}))