mrebuild my repo

Author: Yeah-Kun

.gitattributes

@@ -1 +1,4 @@
 *.tfrecords filter=lfs diff=lfs merge=lfs -text
+DeepLearning/Verification_code_identification/model/new_crack_captcha.model-3000.data-00000-of-00001 filter=lfs diff=lfs merge=lfs -text
+DeepLearning/Verification_code_identification/tmp/events.out.tfevents.1527043331.DESKTOP-6RNLI2A filter=lfs diff=lfs merge=lfs -text
+DeepLearning/Verification_code_identification/model/crack_captcha.model-6000.data-00000-of-00001 filter=lfs diff=lfs merge=lfs -text

DeepLearning/Verification_code_identification/.ipynb_checkpoints/test_unit-checkpoint.ipynb

@@ -0,0 +1,6 @@
+{
+ "cells": [],
+ "metadata": {},
+ "nbformat": 4,
+ "nbformat_minor": 2
+}

DeepLearning/Verification_code_identification/Model.py

@@ -0,0 +1,325 @@
+"""
+	验证码识别模型
+	create by Ian in 2018-4-13 16:57:10
+"""
+from save_image_into_h5py import TransImg
+import h5py
+import numpy as np
+import tensorflow as tf
+from tensorflow.python.framework import ops
+from tensorflow.python import debug as tf_debug
+from cnn_utils import *
+import matplotlib.pyplot as plt
+import scipy
+from PIL import Image
+from scipy import ndimage
+import math
+from tensorflow.python import debug as tfdbg
+
+class CaptchaRecognition(object):
+	"""docstring for CaptchaRecognition
+	"""
+	def __init__(self, text_len, char_set):
+		super(CaptchaRecognition, self).__init__()
+		self.train_path = None
+		self.test_path = None
+		self.X_train = None
+		self.Y_train = None
+		self.X_test = None
+		self.Y_test = None
+		self.text_len = text_len
+		self.char_set = char_set
+		self.char_set_len = len(char_set)
+	
+
+
+	def text2vec(self, text):
+		"""字符串文本转向量
+		"""
+		vector = np.zeros(self.text_len * self.char_set_len) # 字符串向量，也是输出层向量
+
+		def char2pos(c):
+			"""字符转成字符集相应的序号"""
+			for i in range(self.char_set_len):
+				if c == self.char_set[i]:
+					return i
+
+			raise ValueError('No Map')
+
+		for i, c in enumerate(text):
+			idx = i * self.char_set_len + char2pos(c) # 为1的向量序号
+			vector[idx] = 1
+
+		return vector
+
+	def vec2text(self, vec):
+		"""向量转字符串文本
+		"""
+		char_pos = vec.nonzero()[0] #nonzero是numpy的函数，返回数组中非零元素索引值的数组
+		text = []
+		for i, c in enumerate(char_pos):
+			text.append(self.char_set[c])
+
+		return "".join(text)
+
+
+	def load_datasets(self, train_path, test_path):
+		"""加载训练集和测试集，并进行预处理
+		"""
+		self.train_path = train_path
+		self.test_path = test_path
+		train_dataset = h5py.File(train_path, 'r')
+		train_set_x_orig = np.array(train_dataset["train_set"][:])
+		train_set_y_orig = np.array(train_dataset["train_setlabels"][:])
+
+
+		test_dataset = h5py.File(test_path, 'r')
+		test_set_x_orig = np.array(test_dataset["test_set"][:])
+		test_set_y_orig = np.array(test_dataset["test_setlabels"][:])
+
+		#数据归一化和重塑形状
+		self.X_train = train_set_x_orig/255.
+		self.X_test = test_set_x_orig/255.
+		train_set_y_orig = train_set_y_orig.T
+		test_set_y_orig = test_set_y_orig.T
+		self.Y_train = np.zeros((len(train_set_y_orig), 144))
+		self.Y_test = np.zeros((len(test_set_y_orig), 144))
+
+		#将数据集字符串转换为相对应的向量
+		for i in range(len(train_set_y_orig)):
+			self.Y_train[i][0] = i
+			self.Y_train[i] = self.text2vec(train_set_y_orig[i])
+
+		for i in range(len(test_set_y_orig)):
+			self.Y_test[i][0] = i
+			self.Y_test[i] = self.text2vec(test_set_y_orig[i])
+
+		print("X_train shape: " + str(self.X_train.shape))
+		print("Y_train shape: " + str(self.Y_train.shape))
+		print("X_test shape: " + str(self.X_test.shape))
+		print("Y_test shape: " + str(self.Y_test.shape)) 
+
+
+	def create_placeholders(self, n_H0, n_W0, n_C0, n_y):
+		"""创建占位符，存储x和y的输入值
+		"""
+		X = tf.placeholder(tf.float32, name = 'X',shape=(None, n_H0, n_W0, n_C0))
+		Y = tf.placeholder(tf.float32, name = 'Y', shape=(None, n_y))
+		return X, Y
+
+
+	def init_parameters(self):
+		W1 = tf.get_variable("W1", [11,11,3,48], initializer = tf.contrib.layers.xavier_initializer(seed = 1))  
+		b1 = tf.get_variable("b1", [48], initializer = tf.zeros_initializer())  
+		W2 = tf.get_variable("W2", [5,5,48,96], initializer = tf.contrib.layers.xavier_initializer(seed = 1))  
+		b2 = tf.get_variable("b2", [96], initializer = tf.zeros_initializer())
+		W3 = tf.get_variable("W3", [3,3,96,128], initializer = tf.contrib.layers.xavier_initializer(seed = 1))  
+		b3 = tf.get_variable("b3", [128], initializer = tf.zeros_initializer()) 
+
+		parameters = {"W1":W1, "W2":W2, "b1":b1, "b2":b2, "W3":W3, "b3":b3}
+		return parameters
+
+
+
+	def forward_propagation(self, X, parameters):
+		"""定义模型，使用LeNet-5，不过卷积层改用AlexNet的参数
+		"""
+		#keep_prob = 0.8
+		# 两个卷积层
+		W1 = parameters['W1']
+		b1 = parameters['b1']
+		W2 = parameters['W2']
+		b2 = parameters['b2']
+		W3 = parameters['W3']
+		b3 = parameters['b3']
+		with tf.name_scope("conv"):
+			conv1 = tf.nn.relu(tf.nn.bias_add(tf.nn.conv2d(X, W1, strides=[1, 4, 4, 1], padding = 'VALID'), b1))
+			conv1 = tf.nn.max_pool(conv1, ksize=[1, 3, 3, 1], strides=[1, 1, 1, 1], padding='SAME')
+
+			conv2 = tf.nn.relu(tf.nn.bias_add(tf.nn.conv2d(conv1, W2, strides=[1, 1, 1, 1], padding='SAME'), b2))
+			conv2 = tf.nn.max_pool(conv2, ksize=[1, 3, 3, 1], strides=[1, 2, 2, 1], padding='SAME')
+
+			conv3 = tf.nn.relu(tf.nn.bias_add(tf.nn.conv2d(conv2, W3, strides=[1, 1, 1, 1], padding='SAME'), b3))
+			conv3 = tf.nn.max_pool(conv3, ksize=[1, 3, 3, 1], strides=[1, 2, 2, 1], padding='SAME')
+			conv3 = tf.contrib.layers.flatten(conv3)
+
+		with tf.name_scope("output"):
+			w_out = tf.Variable(0.01*tf.random_normal([3*3*128*self.text_len, self.text_len*self.char_set_len]))
+			b_out = tf.Variable(0.1*tf.random_normal([self.text_len*self.char_set_len]))
+			out = tf.add(tf.matmul(conv3, w_out), b_out)
+
+		tf.summary.histogram("W1", W1)
+		tf.summary.histogram("b1", b1)
+		tf.summary.histogram("W2", W2)
+		tf.summary.histogram("b2", b2)
+		tf.summary.histogram("W3", W3)
+		tf.summary.histogram("b3", b3)
+		tf.summary.histogram("w_out", w_out)
+		tf.summary.histogram("b_out", b_out)
+		return out
+
+
+
+	def crack_captcha_cnn(self, X, w_alpha=0.01, b_alpha=0.1):
+	 
+		# 3 conv layer
+		with tf.name_scope("conv"):
+			w_c1 = tf.Variable(w_alpha*tf.random_normal([3, 3, 3, 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')
+	 
+		with tf.name_scope("Fully_connected"):
+			w_d = tf.Variable(w_alpha*tf.random_normal([8*18*64, 1024]))
+			b_d = tf.Variable(b_alpha*tf.random_normal([1024]))
+			dense = tf.contrib.layers.flatten(conv3)
+			dense = tf.nn.relu(tf.add(tf.matmul(dense, w_d), b_d))
+
+	 
+		with tf.name_scope("output"):
+			w_out = tf.Variable(w_alpha*tf.random_normal([1024, self.text_len*self.char_set_len]))
+			b_out = tf.Variable(b_alpha*tf.random_normal([self.text_len*self.char_set_len]))
+			out = tf.add(tf.matmul(dense, w_out), b_out)
+			print(out)
+		tf.summary.histogram("w_out", w_out)
+		tf.summary.histogram("b_out", b_out)
+		#out = tf.nn.softmax(out)
+
+		return out
+
+
+
+	def compute_cost(self, Y_hat, Y):
+		"""损失函数计算
+		"""
+		with tf.name_scope("cost"):
+			#cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits_v2(logits = Y_hat, labels = Y))
+			cost = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits = Y_hat, labels = Y))
+			#cost = tf.reduce_mean(tf.square(Y-Y_hat))
+		tf.summary.scalar("cost", cost)
+		return cost
+
+
+	def tensorflow_model(self, learning_rate = 0.009, num_epochs = 4, minibatch_size = 128, print_cost = True):
+		"""整合各个步骤
+		"""
+		ops.reset_default_graph()
+		(m, n_H0, n_W0, n_C0) = self.X_train.shape # 读取训练集的宽高和channels
+		n_y = self.Y_train.shape[1] # 读取标签预测值
+		seed = 3 # 用于随机数
+		costs = []
+
+		with tf.device('/gpu:0'):
+			X, Y = self.create_placeholders(n_H0, n_W0, n_C0, n_y)
+			parameters = self.init_parameters()
+			Y_hat = self.forward_propagation(X, parameters)
+			#Y_hat = self.crack_captcha_cnn(X)
+			cost = self.compute_cost(Y_hat, Y)
+			optimizer = tf.train.AdamOptimizer(learning_rate = learning_rate).minimize(cost)
+			init = tf.global_variables_initializer()
+
+		saver = tf.train.Saver(parameters) # 保存模型
+		
+		with tf.Session() as sess:
+			sess = tfdbg.LocalCLIDebugWrapperSession(sess) # 被调试器封装的会话
+			sess.add_tensor_filter("has_inf_or_nan", tfdbg.has_inf_or_nan)  # 调试器添加过滤规则
+			writer = tf.summary.FileWriter(r'D:\code\python\DeepLearning\tmp', tf.get_default_graph()) # 将运行概要写入硬盘
+			merged_summary = tf.summary.merge_all() # 将所有的概要打包
+			sess = tf.Session(config=tf.ConfigProto(allow_soft_placement=True, log_device_placement=False)) #log_device_placement：将设备信息展示在控制台
+			sess.run(init)
+			for epoch in range(num_epochs):
+				minibatch_cost = 0
+				num_minibatches = int(m / minibatch_size) # 计算一共有多少mini-batch
+				#seed = seed + 1
+				minibatches = random_mini_batches(self.X_train, self.Y_train, minibatch_size, seed) # 随机混淆数据
+
+				# 从数据集中抽取数据，并进行迭代
+				for minibatch in minibatches:
+					(minibatch_X, minibatch_Y) = minibatch
+					_ , temp_cost, summary, Y_hat_result = sess.run([optimizer, cost, merged_summary, Y_hat], feed_dict={X: minibatch_X, Y: minibatch_Y})
+					writer.add_summary(summary, epoch) # 将数据填入tensorboard
+
+					minibatch_cost += temp_cost / num_minibatches
+
+					print(Y_hat_result)
+					sess.run(tf.Print(Y_hat_result, [Y_hat_result], message="输出层："))
+				
+					#print(Y_hat.eval())
+
+
+				#debugtext = sess.run(tf.Print(cost, [cost], message="输出层：")) # 调试语句
+				#print(debugtext[0] ,debugtext.shape)
+
+				if print_cost == True and epoch % 1 == 0:
+					costs.append(minibatch_cost)
+					if print_cost == True and epoch % 1 == 0:  
+						print ("Cost after epoch %i: %f" % (epoch, minibatch_cost))
+						if  epoch % 50 == 0:
+							saver.save(sess, './model/model.ckpt', global_step=epoch) # 保存模型
+
+
+		plt.plot(np.squeeze(costs))
+		plt.ylabel('cost')  
+		plt.xlabel('iterations (per tens)')  
+		plt.title("Learning rate =" + str(learning_rate))  
+		plt.show()
+
+		parameters = sess.run(parameters)  
+		print ("Parameters have been trained!")
+
+		correct_prediction = tf.equal(Y_hat, Y) # equal:判断矩阵是否一致
+		accuracy = tf.reduce_mean(tf.cast(correct_prediction, bool))
+		print ("Train Accuracy:", accuracy.eval(session=sess, feed_dict={X: self.X_train[:300], Y: self.Y_train[:300]}))
+		print ("Test Accuracy:", accuracy.eval(session=sess, feed_dict={X: self.X_test, Y: self.Y_test}))
+		return parameters
+
+
+
+	def predict(self):
+		"""
+		"""
+		n_y = self.Y_test.shape[1] # 读取标签预测值
+		
+		with tf.Session() as sess:
+			(m, n_H0, n_W0, n_C0) = self.X_test.shape # 读取训练集的宽高和channels
+			print(self.X_test.shape)
+			X, Y = self.create_placeholders(n_H0, n_W0, n_C0, n_y)
+			parameters = self.init_parameters()
+			Y_hat = self.forward_propagation(X, parameters)
+
+		tf.reset_default_graph()
+		saver = tf.train.import_meta_graph('./model/model.ckpt-2100.meta')
+		with tf.Session() as sess:
+			# 加载模型
+			saver.restore(sess, tf.train.latest_checkpoint("./model")) 
+			text_list = sess.run(Y_hat, feed_dict={X: self.X_test})
+			print(type(text_list))
+			print(text_list)
+			print(text_list.shape)
+
+			
+
+
+
+
+if __name__ == '__main__':
+	train_path = "./datasets/train_vcode.h5"
+	test_path = "./datasets/test_vcode.h5"
+	char_set = ['0', '1', '2', '3', '4', '5', '6', '7', '8', '9', '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']
+	c = CaptchaRecognition(4, char_set)
+	c.load_datasets(train_path, test_path)
+	#c.tensorflow_model(learning_rate = 0.001, num_epochs= 50, minibatch_size=128)
+	c.predict()