鏈接：https://pan.baidu.com/s/1aLFV-QovCeig4bGaS8U_8w ,提取碼：5u9h
#-*- coding:utf-8 -*-
from urllib.request import urlretrieve
import time, random, os
class Discuz():
def __init__(self):
# Discuz驗證碼生成圖片地址
self.url = 'http://cuijiahua.com/tutrial/discuz/index.php?label=' #地址失效，請更換！！！
def random_captcha_text(self, captcha_size = 4):
""" 驗證碼一般都無視大小寫；驗證碼長度4個字符 Parameters: captcha_size:驗證碼長度 Returns: captcha_text:驗證碼字符串 """
number = ['0','1','2','3','4','5','6','7','8','9']
alphabet = ['a','b','c','d','e','f','g','h','i','j','k','l','m','n','o','p','q','r','s','t','u','v','w','x','y','z']
char_set = number + alphabet
captcha_text = []
for i in range(captcha_size):
c = random.choice(char_set)
captcha_text.append(c)
captcha_text = ''.join(captcha_text)
return captcha_text
def download_discuz(self, nums = 50000):
""" 下載驗證碼圖片 Parameters: nums:下載的驗證碼圖片數量 """
dirname = './pic'
if dirname not in os.listdir():
os.mkdir(dirname)
for i in range(nums):
label = self.random_captcha_text()
print('第%d張圖片:%s下載' % (i + 1,label))
urlretrieve(url = self.url + label, filename = dirname + '/' + label + '.jpg')
# 請至少加200ms延時，避免給我的服務器造成過多的壓力，如發現影響服務器正常工作，我會關閉此功能。
# 你好我也好，大家好才是真的好！
time.sleep(0.1)
print('恭喜圖片下載完成！')
if __name__ == '__main__':
dz = Discuz()
dz.download_discuz()
#-*- coding:utf-8 -*-
import tensorflow as tf
import matplotlib.pyplot as plt
import numpy as np
import os, random, cv2
class Discuz():
def __init__(self):
# 數據集路徑
self.data_path = './pic/'
# 寫到指定的磁盤路徑中
self.log_dir = './logs/'
# 數據集圖片大小
self.width = 30
self.heigth = 100
# 最大迭代次數
self.max_steps = 100
# 讀取數據集
self.test_imgs, self.test_labels, self.train_imgs, self.train_labels = self.get_imgs()
# 訓練集大小
self.train_size = len(self.train_imgs)
# 測試集大小
self.test_size = len(self.test_imgs)
# 每次獲得batch_size大小的當前訓練集指針
self.train_ptr = 0
# 每次獲取batch_size大小的當前測試集指針
self.test_ptr = 0
# 字符字典大小:0-9 a-z A-Z _(驗證碼如果小於4，用_補齊) 一共63個字符
self.char_set_len = 63
# 驗證碼最長的長度為4
self.max_captcha = 4
# 輸入數據X占位符
self.X = tf.placeholder(tf.float32, [None, self.heigth*self.width])
# 輸入數據Y占位符
self.Y = tf.placeholder(tf.float32, [None, self.char_set_len*self.max_captcha])
# keepout占位符
self.keep_prob = tf.placeholder(tf.float32)
def get_imgs(self, rate = 0.2):
# 讀取圖片
imgs = os.listdir(self.data_path)
# 打亂圖片順序
random.shuffle(imgs)
# 數據集總共個數
imgs_num = len(imgs)
# 按照比例求出測試集個數
test_num = int(imgs_num * rate / (1 + rate))
# 測試集
test_imgs = imgs[:test_num]
# 根據文件名獲取測試集標簽
test_labels = list(map(lambda x: x.split('.')[0], test_imgs))
# 訓練集
train_imgs = imgs[test_num:]
# 根據文件名獲取訓練集標簽
train_labels = list(map(lambda x: x.split('.')[0], train_imgs))
return test_imgs, test_labels, train_imgs, train_labels
def get_next_batch(self, train_flag=True, batch_size=100):
# 從訓練集獲取數據
if train_flag == True:
if (batch_size + self.train_ptr) < self.train_size:
trains = self.train_imgs[self.train_ptr:(self.train_ptr + batch_size)]
labels = self.train_labels[self.train_ptr:(self.train_ptr + batch_size)]
self.train_ptr += batch_size
else:
new_ptr = (self.train_ptr + batch_size) % self.train_size
trains = self.train_imgs[self.train_ptr:] + self.train_imgs[:new_ptr]
labels = self.train_labels[self.train_ptr:] + self.train_labels[:new_ptr]
self.train_ptr = new_ptr
batch_x = np.zeros([batch_size, self.heigth*self.width])
batch_y = np.zeros([batch_size, self.max_captcha*self.char_set_len])
for index, train in enumerate(trains):
img = np.mean(cv2.imread(self.data_path + train), -1)
# 將多維降維1維
batch_x[index,:] = img.flatten() / 255
for index, label in enumerate(labels):
batch_y[index,:] = self.text2vec(label)
# 從測試集獲取數據
else:
if (batch_size + self.test_ptr) < self.test_size:
tests = self.test_imgs[self.test_ptr:(self.test_ptr + batch_size)]
labels = self.test_labels[self.test_ptr:(self.test_ptr + batch_size)]
self.test_ptr += batch_size
else:
new_ptr = (self.test_ptr + batch_size) % self.test_size
tests = self.test_imgs[self.test_ptr:] + self.test_imgs[:new_ptr]
labels = self.test_labels[self.test_ptr:] + self.test_labels[:new_ptr]
self.test_ptr = new_ptr
batch_x = np.zeros([batch_size, self.heigth*self.width])
batch_y = np.zeros([batch_size, self.max_captcha*self.char_set_len])
for index, test in enumerate(tests):
img = np.mean(cv2.imread(self.data_path + test), -1)
# 將多維降維1維
batch_x[index,:] = img.flatten() / 255
for index, label in enumerate(labels):
batch_y[index,:] = self.text2vec(label)
return batch_x, batch_y
def text2vec(self, text):
""" 文本轉向量 Parameters: text:文本 Returns: vector:向量 """
if len(text) > 4:
raise ValueError('驗證碼最長4個字符')
vector = np.zeros(4 * self.char_set_len)
def char2pos(c):
if c =='_':
k = 62
return k
k = ord(c) - 48
if k > 9:
k = ord(c) - 55
if k > 35:
k = ord(c) - 61
if k > 61:
raise ValueError('No Map')
return k
for i, c in enumerate(text):
idx = i * self.char_set_len + char2pos(c)
vector[idx] = 1
return vector
def vec2text(self, vec):
""" 向量轉文本 Parameters: vec:向量 Returns: 文本 """
char_pos = vec.nonzero()[0]
text = []
for i, c in enumerate(char_pos):
char_at_pos = i #c/63
char_idx = c % self.char_set_len
if char_idx < 10:
char_code = char_idx + ord('0')
elif char_idx < 36:
char_code = char_idx - 10 + ord('A')
elif char_idx < 62:
char_code = char_idx - 36 + ord('a')
elif char_idx == 62:
char_code = ord('_')
else:
raise ValueError('error')
text.append(chr(char_code))
return "".join(text)
def crack_captcha_cnn(self, w_alpha=0.01, b_alpha=0.1):
x = tf.reshape(self.X, shape=[-1, self.heigth, self.width, 1])
# 卷積的filter:一個Tensor。數據維度是四維[filter_height, filter_width, in_channels, out_channels]
# 具體含義是[卷積核的高度, 卷積核的寬度, 圖像通道數, 卷積核個數]
w_c1 = tf.Variable(w_alpha*tf.random_normal([3, 3, 1, 32]))
b_c1 = tf.Variable(b_alpha*tf.random_normal([32]))
conv1 = tf.nn.relu(tf.nn.bias_add(tf.nn.conv2d(x, w_c1, strides=[1, 1, 1, 1], padding='SAME'), b_c1))
conv1 = tf.nn.max_pool(conv1, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')
w_c2 = tf.Variable(w_alpha*tf.random_normal([3, 3, 32, 64]))
b_c2 = tf.Variable(b_alpha*tf.random_normal([64]))
conv2 = tf.nn.relu(tf.nn.bias_add(tf.nn.conv2d(conv1, w_c2, strides=[1, 1, 1, 1], padding='SAME'), b_c2))
conv2 = tf.nn.max_pool(conv2, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')
w_c3 = tf.Variable(w_alpha*tf.random_normal([3, 3, 64, 64]))
b_c3 = tf.Variable(b_alpha*tf.random_normal([64]))
conv3 = tf.nn.relu(tf.nn.bias_add(tf.nn.conv2d(conv2, w_c3, strides=[1, 1, 1, 1], padding='SAME'), b_c3))
conv3 = tf.nn.max_pool(conv3, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')
w_d = tf.Variable(w_alpha*tf.random_normal([4*13*64, 1024]))
b_d = tf.Variable(b_alpha*tf.random_normal([1024]))
dense = tf.reshape(conv3, [-1, w_d.get_shape().as_list()[0]])
dense = tf.nn.relu(tf.add(tf.matmul(dense, w_d), b_d))
dense = tf.nn.dropout(dense, self.keep_prob)
w_out = tf.Variable(w_alpha*tf.random_normal([1024, self.max_captcha*self.char_set_len]))
b_out = tf.Variable(b_alpha*tf.random_normal([self.max_captcha*self.char_set_len]))
out = tf.add(tf.matmul(dense, w_out), b_out)
return out
def train_crack_captcha_cnn(self):
output = self.crack_captcha_cnn()
# 創建損失函數
diff = tf.nn.sigmoid_cross_entropy_with_logits(logits=output, labels=self.Y)
loss = tf.reduce_mean(diff)
tf.summary.scalar('loss', loss)
# 使用AdamOptimizer優化器訓練模型，最小化交叉熵損失
optimizer = tf.train.AdamOptimizer(learning_rate=0.001).minimize(loss)
# 計算准確率
y = tf.reshape(output, [-1, self.max_captcha, self.char_set_len])
y_ = tf.reshape(self.Y, [-1, self.max_captcha, self.char_set_len])
correct_pred = tf.equal(tf.argmax(y, 2), tf.argmax(y_, 2))
accuracy = tf.reduce_mean(tf.cast(correct_pred, tf.float32))
tf.summary.scalar('accuracy', accuracy)
merged = tf.summary.merge_all()
with tf.Session() as sess:
# 寫到指定的磁盤路徑中
train_writer = tf.summary.FileWriter(self.log_dir + '/train', sess.graph)
test_writer = tf.summary.FileWriter(self.log_dir + '/test')
sess.run(tf.global_variables_initializer())
# 遍歷self.max_steps次
for i in range(self.max_steps):
# 迭代500次，打亂一下數據集
if i % 20 == 0:
self.test_imgs, self.test_labels, self.train_imgs, self.train_labels = self.get_imgs()
# 每10次，使用測試集，測試一下准確率
if i % 10 == 0:
batch_x_test, batch_y_test = self.get_next_batch(False, 100)
summary, acc = sess.run([merged, accuracy], feed_dict={
self.X: batch_x_test, self.Y: batch_y_test, self.keep_prob: 1})
print('迭代第%d次 accuracy:%f' % (i+1, acc))
test_writer.add_summary(summary, i)
# 如果准確率大於85%，則保存模型並退出。
if acc > 0.85:
train_writer.close()
test_writer.close()
break
# 一直訓練
else:
batch_x, batch_y = self.get_next_batch(True, 100)
loss_value, _ = sess.run([loss, optimizer], feed_dict={
self.X: batch_x, self.Y: batch_y, self.keep_prob: 1})
print('迭代第%d次 loss:%f' % (i+1, loss_value))
curve = sess.run(merged, feed_dict={
self.X: batch_x_test, self.Y: batch_y_test, self.keep_prob: 1})
train_writer.add_summary(curve, i)
train_writer.close()
test_writer.close()
if __name__ == '__main__':
dz = Discuz()
dz.train_crack_captcha_cnn()
