import tensorflow as tf FLAGS = tf.app.flags.FLAGS tf.app.flags.DEFINE_string("captcha_dir", "./tfrecords", "验证码数据路径") tf.app.flags.DEFINE_integer("batch_size", 100, "每批次训练的样本数") tf.app.flags.DEFINE_integer("label_num", 4, "每个样本目标值数量") tf.app.flags.DEFINE_integer("letter_num", 26, "每个目标值取的字母的可能性个数") # 定义一个随机初始化权重函数 def weight_variables(shape): w = tf.Variable(tf.random_normal(shape=shape, mean=0.0, stddev=1.0)) return w # 定义一个随机初始化偏置函数 def bias_variables(shape): b = tf.Variable(tf.constant(0.0, shape=shape)) return b def read_and_decode(): """读取验证码数据""" # 1.构建文件独立 file_queue = tf.train.string_input_producer([FLAGS.captcha_dir]) # 2.构建阅读器，读取文件内容，默认一个样本 reader = tf.TFRecordReader() key, value = reader.read(file_queue) # tfrecords数据需要解析 features = tf.parse_single_example(value, features={ "image":tf.FixedLenFeature([], tf.string), "label":tf.FixedLenFeature([], tf.string), }) # 解码内容，字符串内容 # 1.解析图片特征值 image = tf.decode_raw(features["image"], tf.uint8) # 2.解析目标值 label = tf.decode_raw(features["label"], tf.uint8) # 改变形状 image_reshape = tf.reshape(image, [20, 80, 3]) label_reshape = tf.reshape(label, [4]) # 进行批处理 image_batch, label_batch = tf.train.batch([image_reshape, label_reshape], batch_size=FLAGS.batch_size, num_threads=2, capacity=10) return image_batch, label_batch def fc_model(image): """ 进行预测结果 image [100, 20, 80, 3] """ with tf.variable_scope("model"): # 1。随机初始化权重，偏置 weights = weight_variables([20*80*3, 4*26]) bias = bias_variables([4*26]) # 将图片数据转换成二维 image_reshape = tf.reshape(image, [-1, 20*80*3]) # 进行全连接层矩阵运算 y_predict = tf.matmul(tf.cast(image_reshape, tf.float32), weights) + bias return y_predict def captcharec(): """验证码识别""" # 1.读取验证码数据 image_batch, label_batch = read_and_decode() # 2.通过输入图片的特征数据，建立模型，得出预测结果 # 一层，全连接层进行预测 # matrix [100, 20*80*3]*[20*80*3, 4*26] + [104] = [100, 4*26] y_predict = fc_model(image_batch) # 目标值[100, 4]转换成one-hot编码==>[100, 4, 26] y_true = tf.one_hot(label_batch, depth=FLAGS.letter_num, on_value=1.0, axis=2) # softmax计算，交叉熵损失计算 with tf.variable_scope("soft_cross"): # 求平均交叉熵损失 loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=tf.reshape(y_true, [FLAGS.batch_size, FLAGS.label_num*FLAGS.letter_num]), logits=y_predict)) # 梯度下降优化损失 with tf.variable_scope("optimizer"): # 0.1是学习率，minimize表示求最小损失 train_op = tf.train.GradientDescentOptimizer(0.0001).minimize(loss) # 计算准确率，三维比较 y_predict：[100, 4*26]==>[100, 4, 26] with tf.variable_scope("acc"): equal_list = tf.equal(tf.argmax(y_true, 2), tf.argmax(tf.reshape(y_predict, [FLAGS.batch_size, FLAGS.label_num, FLAGS.letter_num]), 1)) # equal_list None个样本 [1, 0, 1, 1, 0, 0.....] accuracy = tf.reduce_mean(tf.cast(equal_list, tf.float32)) # 定义初始化变量op init_op = tf.global_variables_initializer() # 开启会话 with tf.Session() as sess: sess.run(init_op) # 定义线程协调器和开启线程 coord = tf.train.Coordinator() # 开启线程读取文件 threads = tf.train.start_queue_runners(sess, coord=coord) # 训练数据 for i in range(5000): sess.run(train_op) print("第{}批次的准确率为：{}".format(i, accuracy.eval())) # 回收线程 coord.request_stop() coord.join(threads) return None if __name__ == "__nain__": captcharec()