跟我学算法- tensorflow 卷积神经网络训练验证码

使用captcha.image.Image 生成随机验证码，随机生成的验证码为0到9的数字，验证码有4位数字组成,这是一个自己生成验证码，自己不断训练的模型

使用三层卷积层，三层池化层，二层全连接层来进行组合

第一步：定义生成随机验证码图片

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']
# 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']

def random_captcha_text(char_set=number, captha_size=4):
    captha_texts = []
    for i in range(captha_size):
        # 随机抽取数字，添加到列表中
        captha_texts.append(random.choice(char_set))
    return captha_texts

def gen_captcha_text_and_image():
    image = ImageCaptcha()
    captcha_texts = random_captcha_text()
    # 列表转换为字符串
    captcha_texts = ''.join(captcha_texts)
    # 产生图片
    captcha = image.generate(captcha_texts)
    captcha_image = Image.open(captcha)
    captcha_image = np.array(captcha_image)
    # 返回字符串和图片
    return captcha_texts, captcha_image

第二步： 生成训练样本

# 把彩图转换为灰度图
def convert2gray(image):
    if len(image.shape)> 2:
        grey = np.mean(image, -1)
        return grey
    else:
        return image
# 把文本转换为可用的标签维度是40
def text2vec(text):
    text_len = len(text)
    int(text[0])
    if text_len > MAX_CAPTCHA:
        raise ValueError('验证码最长4个字符')
    vec = np.zeros(MAX_CAPTCHA*CHAR_SET_LEN)
    for index, c in enumerate(text):

        now_index = index * CHAR_SET_LEN + int(c.strip())
        vec[now_index] = 1
    return vec
# 生成训练样本
def get_next_batch(batch_size=128):
    batch_x = np.zeros([batch_size, IMAGE_HEIGHT*IMAGE_WEIGHT])
    batch_y = np.zeros([batch_size, MAX_CAPTCHA*CHAR_SET_LEN])
    # 有时候生成的图像大小不是(60, 160, 3), 重新生成
    def wrap_gen_captcha_text_and_image():
        text, image = gen_captcha_text_and_image()
        while True:
            if image.shape == (60, 160, 3):
                return text, image

    for i in range(batch_size):
        text, image = wrap_gen_captcha_text_and_image()
        image = convert2gray(image)
        # 转换成的一维的灰度图，使得其范围为(0, 1)
        batch_x[i, :] = image.flatten() / 255  # (image.flatten()-128)/128  mean为0
       # 把输入的文本转换为标签类型
        batch_y[i, :] = text2vec(text)

    return batch_x, batch_y

第三步： 定义CNN，这里的CNN为3层卷积，3层池化， 2层全连接

# 定义CNN
def crack_captcha_cnn(w_alpha=0.01, b_alpha=0.1):
    # [-1, IMAGE_HEIGHT, IMAGE_WEIGHT, 1] -1表示batch_size,1表示样本深度，也就是RGB通道的个数 
    x = tf.reshape(X, [-1, IMAGE_HEIGHT, IMAGE_WEIGHT, 1])

    # 创建w_c1和b_c1的初始化变量
    w_c1 = tf.Variable(w_alpha*tf.random_normal([3, 3, 1, 32]))
    b_c1 = tf.Variable(b_alpha*tf.random_normal([32]))
    # 进行卷积操作
    conv1 = tf.nn.relu(tf.nn.bias_add(tf.nn.conv2d(x, w_c1, strides=(1, 1, 1, 1), padding='SAME'), b_c1))
    # 进行池化操作
    conv1 = tf.nn.max_pool(conv1, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')
    conv1 = tf.nn.dropout(conv1, keep_prob)

    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')
    conv2 = tf.nn.dropout(conv2, keep_prob)

    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')
    conv3 = tf.nn.dropout(conv3, keep_prob)

    # 第一个全连接层
    #8*20*64表示conv3的维度， 60/2/2/2 = 8 160/2/2/2=20 
    w_d = tf.Variable(w_alpha * tf.random_normal([8 * 20 * 64, 1024]))
    b_d = tf.Variable(b_alpha * tf.random_normal([1024]))
    dense = tf.reshape(conv3, [-1, w_d.get_shape().as_list()[0]])
    print(tf.matmul(dense, w_d).shape, b_d.shape)
    dense = tf.nn.relu(tf.add(tf.matmul(dense, w_d), b_d))
    dense = tf.nn.dropout(dense, keep_prob)

    # 第二个全连接层， 不需要激活层
    w_out = tf.Variable(w_alpha * tf.random_normal([1024, MAX_CAPTCHA*CHAR_SET_LEN]))
    b_out = tf.Variable(b_alpha * tf.random_normal([MAX_CAPTCHA*CHAR_SET_LEN]))

    out = tf.add(tf.matmul(dense, w_out), b_out)
    return out

第四步： 定义训练CNN函数

def train_crack_captcha_cnn():
    output = crack_captcha_cnn()
    loss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=output, labels=Y))
    optimizer = tf.train.AdamOptimizer(learning_rate=0.001).minimize(loss)
    predict = tf.reshape(output, [-1, MAX_CAPTCHA, CHAR_SET_LEN])
    max_idx_p = tf.argmax(predict, 2)
    max_idx_l = tf.argmax(tf.reshape(Y, [-1, MAX_CAPTCHA, CHAR_SET_LEN]), 2)
    correct_pred = tf.equal(max_idx_p, max_idx_l)
    accr = tf.reduce_mean(tf.cast(correct_pred, tf.float32))

    saver = tf.train.Saver()
    with tf.Session() as sess:
        #变量初始化
        sess.run(tf.global_variables_initializer())
        step = 0
        # 让它一直都训练直到精度大于0.5
        while True:
            # 生成64个样本
            batch_x, batch_y = get_next_batch(batch_size=64)
            __, _loss = sess.run([optimizer, loss], feed_dict={X:batch_x, Y:batch_y, keep_prob: 0.75})
            print(step, _loss)

            # 每一百次循环计算一次返回值
            if step%100 == 0 :
                batch_text_x, batch_text_y = get_next_batch(batch_size=128)
                acc = sess.run(accr, feed_dict={X:batch_text_x, Y:batch_text_y, keep_prob:1.})
                print(acc)
                # 如果准确率大于0.5就保存模型
                if acc > 0.5:
                    saver.save(sess, '.model/crack_captcha/model')
                    break
            step += 1

第五步： 定义训练好后的预测模型

# 用于训练好后的模型进行预测
def crack_captcha(captcha_image):
    
    output = crack_captcha_cnn()
    # 初始化保存数据
    saver = tf.train.Saver()
    with tf.Session() as sess:
        # 重新加载sess
        saver.restore(sess, '.model/crack_captcha/model')
        
        predict = tf.reshape(output, [-1, MAX_CAPTCHA, CHAR_SET_LEN], 2)
        # 获得CNN之后的结果
        text_list = sess.run(predict, feed_dict={X:[captcha_image], keep_prob:1})
        # 让输出结果变成一个列表
        text = text_list[0].tolist()
        return text

第六步：主要函数用来进行训练，或者测试

if __name__ == '__main__':
    #获得文本和图片
    train = 0
    # 当train=0时进行训练
    if train==0:
        number = ['0', '1', '2', '3', '4', '5', '6', '7', '8', '9']
        text, image = gen_captcha_text_and_image()

        IMAGE_HEIGHT = 60
        IMAGE_WEIGHT = 160
        MAX_CAPTCHA = len(text)
        print('验证码文本最长字符数', MAX_CAPTCHA)
        char_set = number
        CHAR_SET_LEN = len(char_set)

        X = tf.placeholder(tf.float32, [None, IMAGE_HEIGHT*IMAGE_WEIGHT])
        Y = tf.placeholder(tf.float32, [None, MAX_CAPTCHA*CHAR_SET_LEN])
        keep_prob = tf.placeholder(tf.float32)

        train_crack_captcha_cnn()
    # 当trian=1时进行测试    
    elif train == 1:
        text, image = gen_captcha_text_and_image()
        # 将模型转换为灰度图以后再进行测试
        image = convert2gray(image)
        image = image.flatten() / 255
        IMAGE_HEIGHT = 60
        IMAGE_WEIGHT = 160
        MAX_CAPTCHA = len(text)
        print('验证码文本最长字符数', MAX_CAPTCHA)
        char_set = number
        CHAR_SET_LEN = len(char_set)
        X = tf.placeholder(tf.float32, [None, IMAGE_HEIGHT * IMAGE_WEIGHT])
        Y = tf.placeholder(tf.float32, [None, MAX_CAPTCHA * CHAR_SET_LEN])
        keep_prob = tf.placeholder(tf.float32)
        pred_text = crack_captcha(image)
        print('真实值', text, '测试值', pred_text)