makeCnnModel.py

import tensorflow as tf
import numpy as np
from PIL import Image
import os,random,time
# 参考链接
# https://blog.csdn.net/sushiqian/article/details/78305340#reply
CAPTCHA_IMAGE_PATH='./images/'
#验证码路径

CAPTCHA_IMAGE_WIDTH=160
#验证码宽度

CAPTCHA_IMAGE_HERGHT=60
#验证码高度

CHAR_SET_LEN=10
CAPTCHA_LEN=4

TRAIN_IMAGE_PERCENT=0.6#60%的验证码图片放入训练集中

TRAIN_IMAGE_NAME=[]#训练集，用于训练的验证验证码图片的文件名

VALIDATTON_IMAGE_NAME=[]#验证集，用于模型验证的验证码图pain的文件名

MODEL_SAVE_PATH='./models/'#存放训练好的模型的路径

def get_image_file_name(imagePath=CAPTCHA_IMAGE_PATH):
    fileName=[]
    total=0
    for filePath in os.listdir(imagePath):
        captcha_name=filePath.split('/')[-1]
        fileName.append(captcha_name)
        total+=1
    return fileName,total


#将验证码转换训练用的标签向量，维数是40
#例如，如果验证码是'0296',则对应的标签是
# [[1 0 0 0 0 0 0 0 0 0],
#  [0 0 1 0 0 0 0 0 0 0],
#  [0 0 0 0 0 0 0 0 0 1],
#  [0 0 0 0 0 0 1 0 0 0]]
def name2label(name):
    #4*10 4行10列矩阵
    label=np.zeros(CAPTCHA_LEN*CHAR_SET_LEN)
    for i,c in enumerate(name):
        idx=i*CHAR_SET_LEN+int(c)
        # idx=i*CHAR_SET_LEN+ord(c)-ord('0')
        label[idx]=1
    return label




#生成一个训练batch
def get_next_batch(batchSize=32,trainOrTest='train',step=0):
                        # 32 , 160*60
    batch_data=np.zeros([batchSize,CAPTCHA_IMAGE_WIDTH*CAPTCHA_IMAGE_HERGHT])
                        #32行   4*10
    batch_label=np.zeros([batchSize,CAPTCHA_LEN*CHAR_SET_LEN])
    fileNmaeList=TRAIN_IMAGE_NAME#需要从这里传如图片路径参数，此参数从 get_image_file_name()函数的到
    if trainOrTest=='validate':
        fileNmaeList=VALIDATTON_IMAGE_NAME
    totalNumber=len(fileNmaeList)
    indexStart=step*batchSize
    for i in range(batchSize):
        # print(i,indexStart,totalNumber)
        index=(i+indexStart)%totalNumber#求余数都忘了13%190000 余13
        name=fileNmaeList[index]
        img_data,img_label=get_data_and_label(name)
        batch_data[i,:]=img_data
        batch_label[i,:]=img_label
        # print(len(batch_data[0]),'-------\n-------',batch_label[0])
        # quit()
    return batch_data,batch_label
# print(get_image_file_name())


#取得验证码图片的数据以及它的标签
def get_data_and_label(fileName,filePath=CAPTCHA_IMAGE_PATH):
    pathName=os.path.join(filePath,fileName)
    img=Image.open(pathName)

    img=img.convert("L")#转化为灰度图
    img_array=np.array(img)#转化成图片数组
    image_data=img_array.flatten()/255 #将多维数组转化成一维数组，值除以255
    image_label=name2label(fileName[0:CAPTCHA_LEN])
    return image_data,image_label

# TRAIN_IMAGE_NAME,labelNum=get_image_file_name()
# get_next_batch(batchSize=32,trainOrTest='train',step=1)#需要传入一个图片位置列表 上面有方法可以拿到


#构建卷积神经网络并训练
def train_data_with_CNN():
    #初始化权重值
    def weight_variable(shape,name='weight'):
        init=tf.truncated_normal(shape,stddev=0.1)
        var=tf.Variable(initial_value=init,name=name)
        return var

    #初始化偏置
    def bias_variable(shape,name='bias'):
        init=tf.constant(0.1,shape=shape)
        var=tf.Variable(init,name=name)
        return var

    #卷积
    def conv2d(x,W,name='conv2d'):
        return tf.nn.conv2d(x,W,strides=[1,1,1,1],padding='SAME',name=name)

    #池化
    def max_pool_2X2(x,name='maxpool'):
        return tf.nn.max_pool(x,ksize=[1,2,2,1],strides=[1,2,2,1],padding='SAME',name=name)

    #注意点X的name ，测试时会用到
    #张量,None未知数量 剩下的是纬度
    X=tf.placeholder(tf.float32,[None,CAPTCHA_IMAGE_WIDTH*CAPTCHA_IMAGE_HERGHT],name='data-input')
    #                                     4*10的长度还是[---40个元素----]
    Y=tf.placeholder(tf.float32,[None,CAPTCHA_LEN*CHAR_SET_LEN],name='label-input')

    x_input=tf.reshape(X,[-1,CAPTCHA_IMAGE_HERGHT,CAPTCHA_IMAGE_WIDTH,1],name='x-input')

    #dropout,防止过拟合
    #注意keep_prob的name，在测试model时会用到
    keep_prob=tf.placeholder(tf.float32,name='keep-prob')

    #第一层卷积
    W_conv1=weight_variable([5,5,1,32],'W_conv1')
    B_conv1=bias_variable([32],'B_conv1')
    conv1=tf.nn.relu(conv2d(x_input,W_conv1,'conv1')+B_conv1)
    conv1=max_pool_2X2(conv1,'conv1-pool')
    conv1=tf.nn.dropout(conv1,keep_prob)

    #第二层卷积
    W_conv2=weight_variable([5,5,32,64],'W_conv2')
    B_conv2=bias_variable([64],'B_conv2')
    conv2=tf.nn.relu(conv2d(conv1,W_conv2,'conv2')+B_conv2)
    conv2=max_pool_2X2(conv2,'conv2-pool')
    conv2=tf.nn.dropout(conv2,keep_prob)

    #第三层卷积
    W_conv3=weight_variable([5,5,64,64],'W_conv3')
    B_conv3=bias_variable([64],'B_conv3')
    conv3=tf.nn.relu(conv2d(conv2,W_conv3,'conv3')+B_conv3)
    conv3=tf.nn.dropout(conv3,keep_prob)

    #全链接层
    #每次池化后，图片的宽度和高度缩小为原来的一半，160 60 /8 =20 7.5
    # 经过上面的三次池化，宽度和高度均缩小8倍

    W_fc1=weight_variable([20*8*64,1024],'W_fcl')
    B_fc1=bias_variable([1024],'B_fc1')
    fc1=tf.reshape(conv3,[-1,20*8*64])
    fc1=tf.nn.relu(tf.add(tf.matmul(fc1,W_fc1),B_fc1))
    fc1=tf.nn.dropout(fc1,keep_prob)

    #输出层                         4*10
    W_fc2=weight_variable([1024,CAPTCHA_LEN*CHAR_SET_LEN],'W_fc2')
    B_fc2=bias_variable([CAPTCHA_LEN*CHAR_SET_LEN],'B_fc2')
    output=tf.add(tf.matmul(fc1,W_fc2),B_fc2,'output')

    loss=tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(labels=Y,logits=output))
    optimizer=tf.train.AdamOptimizer(0.5).minimize(loss)

    predict=tf.reshape(output,[-1,CAPTCHA_LEN,CHAR_SET_LEN],name='predict')
    labels=tf.reshape(Y,[-1,CAPTCHA_LEN,CHAR_SET_LEN],name='labels')

    #预测结果
    #请注意 predict_max_idx 的name ，在测试model时会用到它

    predict_max_idx=tf.argmax(predict,axis=2,name='predict_max_ids')
    labels_max_ids=tf.argmax(labels,axis=2,name='labels_max_ids')
    predict_correct_vec=tf.equal(predict_max_idx,labels_max_ids)
    #tf.cast转化格式 此处为转换成float32
    accuracy=tf.reduce_mean(tf.cast(predict_correct_vec,tf.float32))

    saver=tf.train.Saver()
    with tf.Session() as sess:
        sess.run(tf.global_variables_initializer())
        steps=0
        for epoch in range(6000):#训练600次
            train_data,train_label=get_next_batch(64,'train',step=steps)
            sess.run(optimizer,feed_dict={X:train_data,Y:train_label,keep_prob:0.75})
            if steps%100==0:
                test_data,test_label=get_next_batch(100,'validate',steps)
                acc=sess.run(accuracy,feed_dict={X:test_data,Y:test_label,keep_prob:1.0})
                print("step=%d,accuracy=%f"%(steps,acc))
                if acc>0.99:
                    saver.save(MODEL_SAVE_PATH+"crack_captcha.model",global_step=steps)
                    break
        steps+=1

if __name__ == '__main__':
    image_filename_list,total=get_image_file_name(CAPTCHA_IMAGE_PATH)
    random.seed(time.time())

    #打乱顺序
    random.shuffle(image_filename_list)
    trainImageNumber=int(total*TRAIN_IMAGE_PERCENT)#训练集比例 0.6

    #分成训练集
    TRAIN_IMAGE_NAME=image_filename_list[:trainImageNumber]

    #验证集
    VALIDATTON_IMAGE_NAME=image_filename_list[trainImageNumber:]
    train_data_with_CNN()
    print('Train finished')
    tf.train.GradientDescentOptimizer()#梯度下降优化器
    tf.train.AdamOptimizer()
    tf.train.MomentumOptimizer()