从爬虫构建数据集到CNN模型的验证码识别，一步一步搭建基于Pyt

写在前面：这个程序不是一个人能在短时间内完成的，感谢达纳，王哥的支持帮助。也感谢小平老师，没有压迫，就没有项目。

简介：这是一篇很硬核的Blog, 有一定Python基础的童鞋方能看懂，本程序的主要内容是首先通过Python的Selenium爬虫对12306的图形验证码进行批量爬取，然后通过Pillow对图片的尺寸进行剪裁。将剪裁好的图片分为文字和图形部分并将其打上标签，分别构建卷积神经网络模型(CNN)对图片进行学习，最后通过一系列的整合，用于12306自动抢票，并将成功信息通过邮件方式发送给用户。 机器学习和爬虫 Python

1、搭建12306爬虫程序 12306官网

#利用Selenium批量爬取验证码
#建立存储目录
import time
import os
from lxml import etree
from urllib.request import urlopen
file_dir=r'E:\12306'#设定存储目录
if not os.path.exists(file_dir):
    os.mkdir(file_dir)#判断文件夹是否已经存在
#导入selenium库
import selenium.webdriver as wb
#将谷歌游览器设置为自动化处理
br=wb.Chrome()
#进入到12306官网
br.get('https://kyfw.12306.cn/otn/resources/login.html')
#获取验证码按钮
time.sleep(5)#程序休眠，防止IP被封杀
button=br.find_element_by_xpath('/html/body/div[2]/div[2]/ul/li[2]/a')
time.sleep(3)
button.click()
#获取页面HTML源码病设定循环
i=0
while i<=1000:#设定爬取1000次
    time.sleep(2)
    page=br.page_source
#用xpath语言对目标图片地址进行提取
    html=etree.HTML(page)
    img=html.xpath('//*[@id="J-loginImg"]')[0]
    img_url=img.attrib['src']
#用urllib库去请求网址得到图片的二进制数据
    respond=urlopen(img_url)
    print('done')
    img_bytes=respond.file.read()
#讲读取的图片二进制写入文件并保存
    with open(file_dir+'\\name_1_%d.jpg'%i,'wb') as f:#
        f.write(img_bytes)
        for m in range(20):
            print('*',end='')
            time.sleep(0.01)
        print('完成第%d张图片下载'%(i+1))
    fresh=br.find_element_by_xpath('/html/body/div[2]/div[2]/div[1]/div[2]/div[3]/div/div[3]')
    time.sleep(5)#对验证码网页进行刷新，开启下一轮爬取
    fresh.click()
    i+=1

到目前为止，验证码基本获取完毕，结果如下，我们当时为了更好的结果，爬取了3万多张，结果到后面识别准确率也只能达到90%左右，说明还是不够。 12306验证码

2、图片剪切，将图片剪切为指定大小，供于CNN模型学习

#将图片进行剪切分类
import os
from PIL import Image
import time
char_dir=r'E:\12306_cha'#构造文字部分存储目录
pic_dir=r'E:\12306_pic'#构造图形部分存储目录
ticket_dir=r'E:\12306'#爬虫图片被保存的位置
#判断文件夹是否已经存在，如果没有，新建
if not os.path.exists(char_dir):
    os.mkdir(char_dir)
if not os.path.exists(pic_dir):
    os.mkdir(pic_dir)
#获取文件列表
pic_list=os.listdir(ticket_dir)
for i,num in zip(pic_list,range(len(pic_list))):
    try:
        real=ticket_dir+'\\'+i
        image=Image.open(real)
        i=i.replace('.jpg','')
        cp_ch=image.crop((117,0,230,26))#剪裁文字部分尺寸
        cp_ch.save(char_dir+'\\'+'%s.jpg'%i)#保存文字部分
        cp_pic_1_1=image.crop((3,39,72,109))#剪裁图片部分尺寸
        cp_pic_1_1.save(pic_dir+'\\'+'%s_1_1.jpg'%i)#保存第一张图片，下面同理不在赘述
        cp_pic_1_2=image.crop((75,39,144,109))
        cp_pic_1_2.save(pic_dir+'\\'+'%s_1_2.jpg'%i)
        cp_pic_1_3=image.crop((147,39,216,109))
        cp_pic_1_3.save(pic_dir+'\\'+'%s_1_3.jpg'%i)
        cp_pic_1_4=image.crop((219,39,288,109))
        cp_pic_1_4.save(pic_dir+'\\'+'%s_1_4.jpg'%i)
        cp_pic_2_1=image.crop((3,110,72,180))
        cp_pic_2_1.save(pic_dir+'\\'+'%s_2_1.jpg'%i)
        cp_pic_2_2=image.crop((75,110,144,180))
        cp_pic_2_2.save(pic_dir+'\\'+'%s_2_2.jpg'%i)
        cp_pic_2_3=image.crop((147,110,216,180))
        cp_pic_2_3.save(pic_dir+'\\'+'%s_2_3.jpg'%i)
        cp_pic_2_4=image.crop((219,110,288,180))
        cp_pic_2_4.save(pic_dir+'\\'+'%s_2_4.jpg'%i)
        for n in range(30):
            print('*',end='')
            time.sleep(0.02)
    except:
        print('image error')
        continue
    print('第%d张图片已经处理，还剩%d张'%(num+1,len(pic_list)-num-1))
print('Having done all the pictures')

到这一步，一张验证码图片就被分为1张文字图片和8张图形图片分别保存在各自的文件夹中，效果如下： 文字部分 图形部分

3、接下来一步，就是最让人自闭的打标签，将类别一样的整理到一个文件夹中，我人已经分傻了，能体会从3万张图片中找图片的痛苦吗？不过还是感谢达纳同学分好的小部分数据集，正是这部分数据集用于机器学习，才使得后面3万张被成功分类。效果图如下： 图形分类 文字分类

4、将图片做合适的处理，喂给CNN模型进行学习,模型搭建如下：

首先训练文字部分

#文字部分CNN网络
#原模型修改
# 导入所需模块
from keras.models import Sequential
from keras.layers.normalization import BatchNormalization
from keras.layers.convolutional import Conv2D
from keras.layers.convolutional import MaxPooling2D
from keras.initializers import TruncatedNormal
from keras.layers.core import Activation
from keras.layers.core import Flatten
from keras.layers.core import Dropout
from keras.layers.core import Dense
from keras import backend as K

class SimpleVGGNet:
    @staticmethod
    def build(width, height, depth, classes):
        model = Sequential()
        inputShape = (height, width, depth)
        chanDim = -1

        if K.image_data_format() == "channels_first":
            inputShape = (depth, height, width)
            chanDim = 1

        # CONV => RELU => POOL
        model.add(Conv2D(32, (3, 3), padding="same",
            input_shape=inputShape,kernel_initializer=TruncatedNormal(mean=0.0, stddev=0.01)))
        model.add(Activation("relu"))
        model.add(BatchNormalization(axis=chanDim))
        model.add(MaxPooling2D(pool_size=(3, 3)))#yk注：3*3和2*2（原文）的区别，池化层的尺寸会有影响吗？
        #model.add(Dropout(0.25))#原文是备注掉的，增加的目的是为了增加模型的泛化能力

        # (CONV => RELU) * 2 => POOL
        model.add(Conv2D(64, (3, 3), padding="same",kernel_initializer=TruncatedNormal(mean=0.0, stddev=0.01)))
        model.add(Activation("relu"))
        model.add(BatchNormalization(axis=chanDim))
        model.add(Conv2D(64, (3, 3), padding="same",kernel_initializer=TruncatedNormal(mean=0.0, stddev=0.01)))
        model.add(Activation("relu"))
        model.add(BatchNormalization(axis=chanDim))
        model.add(MaxPooling2D(pool_size=(2, 2)))#yk将2*2修改为3*3
        #model.add(Dropout(0.25))#yk增加了泛化能力

        # (CONV => RELU) * 3 => POOL
        model.add(Conv2D(128, (3, 3), padding="same",kernel_initializer=TruncatedNormal(mean=0.0, stddev=0.01)))
        model.add(Activation("relu"))
        model.add(BatchNormalization(axis=chanDim))
        model.add(Conv2D(128, (3, 3), padding="same",kernel_initializer=TruncatedNormal(mean=0.0, stddev=0.01)))
        model.add(Activation("relu"))
        model.add(BatchNormalization(axis=chanDim))
        model.add(Conv2D(128, (3, 3), padding="same",kernel_initializer=TruncatedNormal(mean=0.0, stddev=0.01)))
        model.add(Activation("relu"))
        model.add(BatchNormalization(axis=chanDim))
        model.add(MaxPooling2D(pool_size=(2, 2)))
        #model.add(Dropout(0.25))

        # FC层
        model.add(Flatten())
        model.add(Dense(1024,kernel_initializer=TruncatedNormal(mean=0.0, stddev=0.01)))#yk将原文的512改为1024
        model.add(Activation("relu"))
        model.add(BatchNormalization())
        #model.add(Dropout(0.25))#yk将0.6改为了0.25

        # softmax 分类
        model.add(Dense(classes,kernel_initializer=TruncatedNormal(mean=0.0, stddev=0.01)))
        model.add(Activation("softmax"))

        return model

CNN网络搭建完毕

# 导入所需工具包
from CNN_net import SimpleVGGNet
from sklearn.preprocessing import LabelBinarizer
from sklearn.model_selection import train_test_split
from sklearn.metrics import classification_report
from keras.optimizers import SGD
from keras.preprocessing.image import ImageDataGenerator
import utils_paths
import matplotlib.pyplot as plt
import numpy as np
import argparse
import random
import pickle
import cv2
import os
import PIL.Image as Image

# 读取数据和标签
print("------开始读取数据------")
data = []
labels = []

# 拿到图像数据路径，方便后续读取
imagePaths = sorted(list(utils_paths.list_images(path_1)))#path_1时文字部分的文件路径
random.seed(42)
random.shuffle(imagePaths)

# 遍历读取数据
for imagePath in imagePaths:
    # 读取图像数据
    image = Image.open(imagePath)
    image =np.array(image)
    image = cv2.resize(image, (96, 96))#修改
    data.append(image)
    # 读取标签
    label = imagePath.split(os.path.sep)[-2]
    labels.append(label)

# 对图像数据做scale操作
data = np.array(data, dtype="float") / 255.0
labels = np.array(labels)

# 数据集切分
(trainX, testX, trainY, testY) = train_test_split(data,labels, test_size=0.25, random_state=42)

# 转换标签为one-hot encoding格式
lb = LabelBinarizer()
trainY = lb.fit_transform(trainY)
testY = lb.transform(testY)

# 数据增强处理
aug = ImageDataGenerator(rotation_range=30, width_shift_range=0.1,
    height_shift_range=0.1, shear_range=0.2, zoom_range=0.2,
    horizontal_flip=True, fill_mode="nearest")

# 建立卷积神经网络
model = SimpleVGGNet.build(width=96, height=96, depth=3,classes=len(lb.classes_))

# 设置初始化超参数
INIT_LR = 0.001
EPOCHS = 100#
BS = 32

# 损失函数，编译模型
print("------准备训练网络------")
opt = SGD(lr=INIT_LR, decay=INIT_LR / EPOCHS)
model.compile(loss="categorical_crossentropy", optimizer=opt,metrics=["accuracy"])

# 训练网络模型
H = model.fit_generator(aug.flow(trainX, trainY, batch_size=BS),
    validation_data=(testX, testY), steps_per_epoch=len(trainX) // BS,
    epochs=EPOCHS)
"""
H = model.fit(trainX, trainY, validation_data=(testX, testY),
    epochs=EPOCHS, batch_size=32)
"""


# 测试
print("------测试网络------")
predictions = model.predict(testX, batch_size=32)
print(classification_report(testY.argmax(axis=1),
    predictions.argmax(axis=1), target_names=lb.classes_))

# 绘制结果曲线
N = np.arange(0, EPOCHS)
plt.style.use("ggplot")
plt.figure()
plt.plot(N, H.history["loss"], label="train_loss")
plt.plot(N, H.history["val_loss"], label="val_loss")
plt.plot(N, H.history["accuracy"], label="train_acc")
plt.plot(N, H.history["val_accuracy"], label="val_acc")
plt.title("Training Loss and Accuracy")
plt.xlabel("Epoch #")
plt.ylabel("Loss/Accuracy")
plt.legend()
plt.savefig(r'path_2\cnn_plot.png')#path_2时学习曲线的保存路径

# 保存模型
print("------正在保存模型------")
model.save(r'path_3\cnn.model')#path_3是模型的保存路径
f = open(r'path_4\cnn_lb.pickle', "wb")#path_4是标签集的保存路径
f.write(pickle.dumps(lb))
f.close()

模型开始进行训练

训练过程 训练结果

可以看到最后的准确率很低，But由于我在其中将Dropout层去除，模型过拟合，导致我得到的结果虽然准确率低，但是识别效果很好（比我的图片识别效果还好）

图形部分CNN模型

#原模型修改
# 导入所需模块
from keras.models import Sequential
from keras.layers.normalization import BatchNormalization
from keras.layers.convolutional import Conv2D
from keras.layers.convolutional import MaxPooling2D
from keras.initializers import TruncatedNormal
from keras.layers.core import Activation
from keras.layers.core import Flatten
from keras.layers.core import Dropout
from keras.layers.core import Dense
from keras import backend as K

class SimpleVGGNet:
    @staticmethod
    def build(width, height, depth, classes):
        model = Sequential()
        inputShape = (height, width, depth)
        chanDim = -1

        if K.image_data_format() == "channels_first":
            inputShape = (depth, height, width)
            chanDim = 1

        # CONV => RELU => POOL
        model.add(Conv2D(32, (3, 3), padding="same",
            input_shape=inputShape,kernel_initializer=TruncatedNormal(mean=0.0, stddev=0.01)))
        model.add(Activation("relu"))
        model.add(BatchNormalization(axis=chanDim))
        model.add(MaxPooling2D(pool_size=(3, 3)))#余坤注：3*3和2*2（原文）的区别，池化层的尺寸会有影响吗？
        model.add(Dropout(0.25))#原文是备注掉的，增加的目的是为了增加模型的泛化能力

        # (CONV => RELU) * 2 => POOL
        model.add(Conv2D(64, (3, 3), padding="same",kernel_initializer=TruncatedNormal(mean=0.0, stddev=0.01)))
        model.add(Activation("relu"))
        model.add(BatchNormalization(axis=chanDim))
        model.add(Conv2D(64, (3, 3), padding="same",kernel_initializer=TruncatedNormal(mean=0.0, stddev=0.01)))
        model.add(Activation("relu"))
        model.add(BatchNormalization(axis=chanDim))
        model.add(MaxPooling2D(pool_size=(3, 3)))#yk将2*2修改为3*3
        model.add(Dropout(0.25))#yk增加了泛化能力

        # (CONV => RELU) * 3 => POOL
        model.add(Conv2D(128, (3, 3), padding="same",kernel_initializer=TruncatedNormal(mean=0.0, stddev=0.01)))
        model.add(Activation("relu"))
        model.add(BatchNormalization(axis=chanDim))
        model.add(Conv2D(128, (3, 3), padding="same",kernel_initializer=TruncatedNormal(mean=0.0, stddev=0.01)))
        model.add(Activation("relu"))
        model.add(BatchNormalization(axis=chanDim))
        model.add(Conv2D(128, (3, 3), padding="same",kernel_initializer=TruncatedNormal(mean=0.0, stddev=0.01)))
        model.add(Activation("relu"))
        model.add(BatchNormalization(axis=chanDim))
        model.add(MaxPooling2D(pool_size=(2, 2)))
        model.add(Dropout(0.25))

        # FC层
        model.add(Flatten())
        model.add(Dense(1024,kernel_initializer=TruncatedNormal(mean=0.0, stddev=0.01)))#yk将原文的512改为1024
        model.add(Activation("relu"))
        model.add(BatchNormalization())
        model.add(Dropout(0.25))#yk将0.6改为了0.5

        # softmax 分类
        model.add(Dense(classes,kernel_initializer=TruncatedNormal(mean=0.0, stddev=0.01)))
        model.add(Activation("softmax"))

        return model

CNN模型搭建完毕

开始对图形部分进行CNN网络训练

# 导入所需工具包
from CNN_net import SimpleVGGNet
from sklearn.preprocessing import LabelBinarizer
from sklearn.model_selection import train_test_split
from sklearn.metrics import classification_report
from keras.optimizers import SGD
from keras.preprocessing.image import ImageDataGenerator
import utils_paths
import matplotlib.pyplot as plt
import numpy as np
import argparse
import random
import pickle
import cv2
import os
import PIL.Image as Image

# 读取数据和标签
print("------开始读取数据------")
data = []
labels = []

# 拿到图像数据路径，方便后续读取
imagePaths =
sorted(list(utils_paths.list_images(path_1)))#path_1是图形的路径
random.seed(42)
random.shuffle(imagePaths)

# 遍历读取数据
for imagePath in imagePaths:
    # 读取图像数据
    image = Image.open(imagePath)
    image =np.array(image)
    image = cv2.resize(image, (80, 80))#修改
    data.append(image)
    # 读取标签
    label = imagePath.split(os.path.sep)[-2]
    labels.append(label)

# 对图像数据做scale操作
data = np.array(data, dtype="float") / 255.0
labels = np.array(labels)

# 数据集切分
(trainX, testX, trainY, testY) = train_test_split(data,labels, test_size=0.25, random_state=42)

# 转换标签为one-hot encoding格式
lb = LabelBinarizer()
trainY = lb.fit_transform(trainY)
testY = lb.transform(testY)

# 数据增强处理
aug = ImageDataGenerator(rotation_range=30, width_shift_range=0.1,
    height_shift_range=0.1, shear_range=0.2, zoom_range=0.2,
    horizontal_flip=True, fill_mode="nearest")

# 建立卷积神经网络
model = SimpleVGGNet.build(width=80, height=80, depth=3,classes=len(lb.classes_))

# 设置初始化超参数
INIT_LR = 0.001
EPOCHS = 100#
BS = 32

# 损失函数，编译模型
print("------准备训练网络------")
opt = SGD(lr=INIT_LR, decay=INIT_LR / EPOCHS)
model.compile(loss="categorical_crossentropy", optimizer=opt,metrics=["accuracy"])

# 训练网络模型
H = model.fit_generator(aug.flow(trainX, trainY, batch_size=BS),
    validation_data=(testX, testY), steps_per_epoch=len(trainX) // BS,
    epochs=EPOCHS)
"""
H = model.fit(trainX, trainY, validation_data=(testX, testY),
    epochs=EPOCHS, batch_size=32)
"""


# 测试
print("------测试网络------")
predictions = model.predict(testX, batch_size=32)
print(classification_report(testY.argmax(axis=1),
    predictions.argmax(axis=1), target_names=lb.classes_))

# 绘制结果曲线
N = np.arange(0, EPOCHS)
plt.style.use("ggplot")
plt.figure()
plt.plot(N, H.history["loss"], label="train_loss")
plt.plot(N, H.history["val_loss"], label="val_loss")
plt.plot(N, H.history["accuracy"], label="train_acc")
plt.plot(N, H.history["val_accuracy"], label="val_acc")
plt.title("Training Loss and Accuracy")
plt.xlabel("Epoch #")
plt.ylabel("Loss/Accuracy")
plt.legend()
plt.savefig(r'path_2\cnn_plot.png')#path_2是学习曲线的存储路径

# 保存模型
print("------正在保存模型------")
model.save(r'path_3\cnn.model')#path_3是你自己的模型存储路径
f = open(r'path_4\cnn_lb.pickle', "wb")#path_4时你自己的标签集存储路径
f.write(pickle.dumps(lb))
f.close()

训练过程,很伤CPU,心疼我的电脑
学习曲线

可以看到最后的准确率很接近1，达到了92%，但我觉得效果还只是一般，还是会有点智障，比如把菠萝识别成啤酒，红豆识别成红枣，没办法，要想准确率更高，必须要有更大的数据集，其次对图片进行合适的处理（我还不是太会）

4.5、这里要提一下啊，我们的做法稍微巧了一点，我们先让机器对1500张左右的图片进行学习，准确率可以达到80%左右。然后我们让这种程度的机器对图片进行分类，最终分完了30000张图片（当然我们最后人工进行挑错了），如下是分类的代码：

# 导入所需工具包
from keras.models import load_model
import argparse
import pickle
import cv2
import PIL.Image as Image
import os
import shutil
import numpy as np
print("------读取模型和标签------")
model = load_model(path)#path是保存模型的路径
lb = pickle.loads(open(path_1, "rb").read())#path_1是保存标签集的路径 
def get_piclist(path):
    return os.listdir(path)
path=r'E:\12306_pic'
pic_list=get_piclist(path)#get picture list
for i in pic_list:
    img_path=path+'\\'+i
    image=np.array(Image.open(img_path))
    image = cv2.resize(image, (80,80))
    image = image.astype("float") / 255.0
    image = image.reshape((1, image.shape[0], image.shape[1],image.shape[2]))
    preds = model.predict(image)
    j = preds.argmax(axis=1)[0]
    label = lb.classes_[j]
    accuracy=int(preds[0][j] * 100)
    print(label+'===>'+str(accuracy))
    if accuracy>10:
        shutil.move(img_path,r'Profile'+'\\'+'%s'%label)#这一步是把图片转移到预测的标签文件中
    else:
        continue

5、进行到这里，基本上绝大部分工作已经做完了，后面的就是把前面的结果和Selenium结合实现自动化抢票,代码如下： Selenium自动化 Chrome宝贝球

#获取时钟的函数
import datetime
def get_time():
    contem=datetime.datetime.now()
    return contem

#发送邮件的函数，以163邮箱为端口
#获取你的邮箱
import smtplib
from email.mime.text import MIMEText
from email.utils import formataddr

my_sender='user'    # 发件人邮箱账号
my_pass = 'password'              # 发件人邮箱密码
def mail(address):
    my_user=address
    ret=True
    try:
        msg=MIMEText('小主，快来12306官网支付您的车票喽！！！','plain','utf-8')
        msg['From']=formataddr(["抢票小助手",my_sender])  # 括号里的对应发件人邮箱昵称、发件人邮箱账号
        msg['To']=formataddr(["FK",my_user])              # 括号里的对应收件人邮箱昵称、收件人邮箱账号
        msg['Subject']="12306提醒"                # 邮件的主题，也可以说是标题

        server=smtplib.SMTP_SSL("smtp.163.com", 465)  # 发件人邮箱中的SMTP服务器，端口是25,这里是163邮箱
        server.login(my_sender, my_pass)  # 括号中对应的是发件人邮箱账号、邮箱密码
        server.sendmail(my_sender,[my_user,],msg.as_string())  # 括号中对应的是发件人邮箱账号、收件人邮箱账号、发送邮件
        server.quit()  # 关闭连接
    except Exception:  # 如果 try 中的语句没有执行，则会执行下面的 ret=False
        ret=False
    return

#获取用户的输入12306账号和密码，并存储为CSV文件以及自己的邮箱
import csv
import pandas as pd
import os
#设置读取或者写入CSV
def get_user():
    path=r'..\User_information\user_information.csv'
    if os.path.exists(path):
        infm=pd.read_csv(path)
        user_name=int(infm.columns[0])
        user_secret=(infm.columns[1])
    else:
        user_name=int(input("Please input your User Name: "))
        user_secret=input("please input your User Secret: ")
        with open(path,'w',newline=None) as f:
            cwriter=csv.writer(f)
            cwriter.writerow([user_name,user_secret])
    return (user_name,user_secret)

#图片识别部分函数
from keras.models import load_model
import argparse
import pickle
import cv2
import PIL.Image as Image
import os
import shutil
import numpy as np
print("------读取模型和标签------")
model_p = load_model(r'..\pic_train\cnn.model')
lb_p = pickle.loads(open(r'..\pic_train\cnn_lb.pickle', "rb").read())
model_c = load_model(r'..\char_train\cnn.model')
lb_c = pickle.loads(open(r'..\char_train\cnn_lb.pickle', "rb").read())
def pic_identify(image):
    image = cv2.resize(image, (80,80))
    image = image.astype("float") / 255.0
    image = image.reshape((1, image.shape[0], image.shape[1],image.shape[2]))
    preds = model_p.predict(image)
    j = preds.argmax(axis=1)[0]
    label = lb_p.classes_[j]
    return label
def char_identify(charc):
    image = cv2.resize(charc, (96, 96))
    image = image.astype("float") / 255.0
    image = image.reshape((1, image.shape[0], image.shape[1],image.shape[2]))
    preds = model_c.predict(image)
    i = preds.argmax(axis=1)[0]
    label = lb_c.classes_[i]
    return label

#登录12306官网函数
import selenium.webdriver as wb
from selenium.webdriver.support import expected_conditions as EC
from selenium.webdriver.support.ui import WebDriverWait as WDW
from selenium.webdriver.common.by import By
import time
from lxml import etree
from urllib.request import urlopen
import PIL.Image as Image
import numpy as np
from selenium.webdriver.common.action_chains import ActionChains
def login(user_name,user_secret,br):
    url='https://kyfw.12306.cn/otn/resources/login.html'
    wait=WDW(br,10)
    br.get(url)
    login_surface=wait.until(EC.presence_of_element_located((By.XPATH,'/html/body/div[2]/div[2]/ul/li[2]/a')))
    time.sleep(5)
    login_surface.click()
    time.sleep(5)
    user_name_button=br.find_element_by_id('J-userName').send_keys(user_name)
    user_secret_button=br.find_element_by_id('J-password').send_keys(user_secret)
    br.maximize_window()
    
def pic_get(br):
    file_dir=r'..\download_pic'
    page=br.page_source
    html=etree.HTML(page)
    img=html.xpath('//*[@id="J-loginImg"]')[0]
    img_url=img.attrib['src']
    respond=urlopen(img_url)
    img_bytes=respond.file.read()
    with open(file_dir+'\\12306.jpg','wb') as f:
        f.write(img_bytes)

def pic_cut():
    pic_path=r'..\download_pic\12306.jpg'
    image=Image.open(pic_path)
    charc=np.array(image.crop((117,0,230,26)))
    pic_1=np.array(image.crop((3,39,72,109)))
    l_1=(1050,350)
    pic_2=np.array(image.crop((75,39,144,109)))
    l_2=(1130,350)
    pic_3=np.array(image.crop((147,39,216,109)))
    l_3=(1200,350)
    pic_4=np.array(image.crop((219,39,288,109)))
    l_4=(1270,350)
    pic_5=np.array(image.crop((3,110,72,180)))
    l_5=(1050,420)
    pic_6=np.array(image.crop((75,110,144,180)))
    l_6=(1130,420)
    pic_7=np.array(image.crop((147,110,216,180)))
    l_7=(1200,420)
    pic_8=np.array(image.crop((219,110,288,180)))
    l_8=(1270,420)
    return ([pic_1,pic_2,pic_3,pic_4,pic_5,pic_6,pic_7,pic_8],[l_1,l_2,l_3,l_4,l_5,l_6,l_7,l_8],charc)
    
def mouse_click(x,y,br):
    ActionChains(br).move_by_offset(x,y).click().perform()
    ActionChains(br).move_by_offset(-x,-y).perform()

#开始抢票,并将上面所有提到的函数集中到此模块中
from get_information import get_user as gu
from login_12306 import *
from identify import *
import time
import datetime
from clock import *
from tqdm import tqdm
from selenium.webdriver.common.keys import Keys
from email_notion import mail
def main():
    #开始记录登录信息
    username,password=gu()
    address=input('请输入您的电子邮箱: ')
    #启动游览器脚本
    br=wb.Chrome()
    #登录12306
    login(username,password,br)
    #等待5秒
    time.sleep(5)
    #下载图片
    pic_get(br)
    #图片剪切
    piclist,location,chara=pic_cut()
    #获取char文字内容
    chara_content=char_identify(chara)
    #识别文字，把机器不易识别的剔除
    while chara_content=='卷尺' or chara_content=='锅铲' or chara_content=='海报' or chara_content=='珊瑚'or chara_content=='棉棒':
        fresh=br.find_element_by_xpath('/html/body/div[2]/div[2]/div[1]/div[2]/div[3]/div/div[3]')
        fresh.click()
        pic_get(br)
        time.sleep(10)
        piclist,location,chara=pic_cut()
        time.sleep(5)
        chara_content=char_identify(chara)
    #选取相对应的图片区
    for i,j in zip(piclist,location):
        if pic_identify(i)==chara_content:
            print(pic_identify(i))
            mouse_click(j[0],j[1],br)
    login_button=br.find_element_by_id('J-login')
    verify=input('How about the choice：')
    if verify=='y':
        login_button.click()
    #选取买票，仅支持单程
    time.sleep(10)
    dan=br.find_element_by_xpath('//*[@id="J-chepiao"]/a')
    dan_v=br.find_element_by_xpath('//*[@id="J-chepiao"]/div/div[1]/ul/li[1]/a')
    dan.click()
    dan_v.click()
    #输入出发地，目的地和出发时间
    fromstButton=br.find_element_by_id('fromStationText')
    tostButton=br.find_element_by_id('toStationText')
    departure=input("请输入出发地:")
    destiney=input("请输入目的地:")
    gotime=input('请输入出发时间(eg.2019-12-3):')
    today=datetime.datetime.now()
    expect=datetime.datetime.strptime(gotime,'%Y-%m-%d')
    year=int(gotime.split('-')[0])
    month=int(gotime.split('-')[1])
    day=int(gotime.split('-')[2])
    fromstButton.click()
    fromstButton.send_keys(departure)
    fromstButton.send_keys(Keys.ENTER)
    time.sleep(1)
    tostButton.click()
    tostButton.send_keys(destiney)
    tostButton.send_keys(Keys.ENTER)
    date=br.find_element_by_id('date_icon_1').click()
    tomonth=br.find_elements_by_xpath('/html/body/div[34]/div[1]/div[2]/div')
    nextmonth=br.find_elements_by_xpath('/html/body/div[34]/div[2]/div[2]/div')
    #date=br.find_element_by_id('date_icon_1')
    #设定循环，如果不是提前一个月的节点，程序会进行休眠
    while expect>today+datetime.timedelta(days=30):
        for i in tqdm(range(3600)):
            time.sleep(1)
        today=get_time()
    if time.localtime().tm_mon==month:
        choice=tomonth[day-1].click()
    else:
        choice=nextmonth[day-1].click()
    #开始查询
    all_button=br.find_element_by_id('train_type_btn_all').click()
    search_button=br.find_element_by_id('query_ticket').click()
    #仅以一个例子为例，太多了设置比较麻烦
    time.sleep(5)
    train_info=etree.HTML(br.page_source)
    #获取列车的始终地
    place=train_info.xpath('//*[@id="train_num_0"]/div[2]/strong')
    #获取列车的初末时间
    getime=train_info.xpath('//*[@id="train_num_0"]/div[3]/strong')
    trainId=train_info.xpath('//*[@id="queryLeftTable"]/tr[1]')[0].attrib['id']
    print("已为您查询到可依靠的列车")
    print('%s ===>> %s   %s ===>> %s'%(place[0].text,place[1].text,getime[0].text,getime[1].text))
    #预定开始
    preorder=br.find_element_by_xpath('//*[@id="%s"]/td[13]/a'%trainId)
    preorder.click()
    time.sleep(2)
    #选择乘客
    passenger=br.find_element_by_id('normalPassenger_0')
    passenger.click()
    #提交订单
    submit=br.find_element_by_id('submitOrder_id').click()
    time.sleep(2)
    #确认
    sure=br.find_element_by_id('qr_submit_id').click()
    mail(address)
    return br
main()

哎，由于我作业缠身，有一部分代码的注释还没有写完，等到闲下来会一一注释清楚，效果图额（主要是我忘了存，再加上调试代码时一天购票三次，已经没机会买票了，就看最终结果吧 最终提示

这就是本篇Blog的主要内容，没办法时间太紧，作业太多，所以有不当之处还请指正，欢迎大家交流学习啊。所有文件程序都已保存到百度网盘 lsbk

补作业 日常结尾