tensorflow1.x及tensorflow2.x不同版本实现验证码识别

import tensorflow as tf
import glob, os,io,sys
import pandas as pd
import numpy as np
os.environ["TF_CPP_MIN_LOG_LEVEL"]='2'def read_pic():
file_list=glob.glob("./code_imgs/*.png")
file_queue=tf.train.string_input_producer(file_list)
reader=tf.WholeFileReader()
filename,image=reader.read(file_queue)
decoded=tf.image.decode_png(image,channels=3)
decoded.set_shape([28,96,3])
image_cast=tf.cast(decoded,tf.float32)
filename_batch,image_batch=tf.train.batch([filename,image_cast],batch_size=40,num_threads=2,capacity=40)
return filename_batch,image_batchdef parse_csv():
csv_data=pd.read_csv('labels.csv',names=['file_num','chars'],index_col='file_num')
labels=[]
for label in csv_data["chars"]:
letter=[]
for word in label:
letter.append(ord(word)-ord('a'))
labels.append(letter)
csv_data['labels']=labels
return csv_datadef filename2label(filenames,csv_data):
labels=[]
for filename in filenames:
file_num="".join(filter(str.isdigit,str(filename)))
target=csv_data.loc[int(file_num),"labels"]
labels.append(target)
return np.array(labels)def create_weights(shape):
return tf.Variable(initial_value=tf.random_normal(shape=shape,stddev=0.01))def create_model(x):
with tf.variable_scope('conv1'):
conv1_weights=create_weights(shape=[5,5,3,32])
conv1_bias=create_weights(shape=[32])
conv1_x=tf.nn.conv2d(input=x,filter=conv1_weights,strides=[1,1,1,1],padding='SAME')+conv1_bias
relu1_x=tf.nn.relu(conv1_x)
pool1_x=tf.nn.max_pool(value=relu1_x,ksize=[1,2,2,1],strides=[1,2,2,1],padding='SAME') with tf.variable_scope('conv2'):
conv2_weights=create_weights(shape=[5,5,32,64])
conv2_bias=create_weights(shape=[64])
conv2_x=tf.nn.conv2d(input=pool1_x,filter=conv2_weights,strides=[1,1,1,1],padding='SAME')+conv2_bias
relu2_x=tf.nn.relu(conv2_x)
pool2_x=tf.nn.max_pool(value=relu2_x,ksize=[1,2,2,1],strides=[1,2,2,1],padding='SAME') with tf.variable_scope('full_connection'):
x_fc=tf.reshape(pool2_x,shape=[-1,7*24*64])
weights_fc=create_weights(shape=[7*24*64,104])
bias_fc=create_weights(shape=[104])
y_predict=tf.matmul(x_fc,weights_fc)+bias_fc return y_predictdef list2text(textlist):
tm=[]
for i in textlist:
tm.append(chr(97+i))
return "".join(tm)def train():
filename,image=read_pic()
csv_data=parse_csv()
x=tf.placeholder(tf.float32,shape=[None,28,96,3])
y_true=tf.placeholder(tf.float32,shape=[None,104])
y_predict=create_model(x)
loss_list=tf.nn.sigmoid_cross_entropy_with_logits(labels=y_true,logits=y_predict)
loss=tf.reduce_mean(loss_list)
optimizer=tf.train.AdamOptimizer(learning_rate=0.001).minimize(loss)#优化损失
equal_list=tf.reduce_all(tf.equal(tf.argmax(tf.reshape(y_predict,shape=[-1,4,26]),axis=2), tf.argmax(tf.reshape(y_true,shape=[-1,4,26]),axis=2)),axis=1)
accurary=tf.reduce_mean(tf.cast(equal_list,tf.float32))
init=tf.global_variables_initializer()
saver=tf.train.Saver()
with tf.Session() as sess:
sess.run(init)
coord=tf.train.Coordinator()
threads=tf.train.start_queue_runners(sess=sess,coord=coord)
try:
for i in range(10000):
filename_val,image_val=sess.run([filename,image])
labels=filename2label(filename_val,csv_data)
labels_value=tf.reshape(tf.one_hot(labels,depth=26),[-1,104]).eval()
_,error,accurary_value=sess.run([optimizer,loss,accurary],feed_dict={x:image_val,y_true:labels_value})
print("The %d Train Result---loss:%f,accurary:%f" % (i+1,error,accurary_value))
if accurary_value>0.99:
saver.save(sess,'model/crack_captcha.model99',global_step=i)
break
except tf.errors.OutOfRangeError:
print("done ,now let's kill all threads")
finally:
coord.request_stop()
print("all threads ask stop")
coord.join(threads)
print("all thread stopped")def crackcaptcha():
truetext=[]
with open('a.txt','r') as f:
for filename in f.readlines():
truetext.append(filename.strip('\r\n'))
dis='False'
goodnum=0 x=tf.placeholder(tf.float32,shape=[None,28,96,3])
y_true=tf.placeholder(tf.float32,shape=[None,104])
keep_prob = tf.placeholder(tf.float32)
y_predict=create_model(x)
saver=tf.train.Saver()
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
checkpoint=tf.train.get_checkpoint_state('model')
if checkpoint and checkpoint.model_checkpoint_path:
saver.restore(sess,checkpoint.model_checkpoint_path)
print("successfully loaded:",checkpoint.model_checkpoint_path)
else:
print("Could not found model files") for i in range(1,201):
image=tf.read_file('crackimgs/'+str(i)+'.png')
decoded=tf.image.decode_png(image,channels=3)
decoded.set_shape([28,96,3])
decoded_val=sess.run(decoded)
image=np.array(decoded_val) predict=tf.argmax(tf.reshape(y_predict,[-1,4,26]),2)
outtext=sess.run(predict,feed_dict={x:[image],keep_prob:1})
cracktext=list2text(outtext[0].tolist())
if cracktext==truetext[i-1]:
goodnum+=1
dis='True'
else:
dis='False'
print('The {} Image Content is:{},Your Crack Word is :{},Result:{}'.format(i,truetext[i-1],cracktext,dis))
print('The End accurary is:{}%'.format(goodnum/200*100))if __name__=='__main__':
train()
crackcaptcha()

np_resource = np.dtype([("resource", np.ubyte, 1)])修改为：np_resource = np.dtype([("resource", np.ubyte, (1,))])