# -*- coding: utf-8 -*-
# http://ju.outofmemory.cn/entry/344970 //layers
# https://my.oschina.net/u/876354/blog/1637819
# https://www.zhihu.com/question/49039504
import os
import tensorflow as tf
import 数据处理_hss as data_hss
import shutil
import numpy as np
# import 模型2sxc as md1
import time
def conv1d_relu(X,filters,kernel_size,strides):
X_change = tf.layers.conv1d(X, filters, kernel_size, strides, 'same', use_bias=True, activation=None)
X_change = tf.nn.relu(X_change)
return X_change
if __name__ == '__main__':
start = time.time()
file_1 = r'D:\hss_python_work\resnet_cnn\数据集\N' #文件夹
file_2 = r'D:\hss_python_work\resnet_cnn\数据集\S'
file_3 = r'D:\hss_python_work\resnet_cnn\数据集\T'
file_4 = r'D:\hss_python_work\resnet_cnn\数据集\V'
file_5 = r'D:\hss_python_work\resnet_cnn\数据集\X'
file_NO_X = r'D:\hss_python_work\resnet_cnn\数据集\非X'
data_all_1,label_all_1 = data_hss.data_analysis(file_NO_X,label_def = [0,1],proportion = 800) # 先定义标签
data_all_2,label_all_2 = data_hss.data_analysis(file_5,label_def = [1,0],proportion = 50) # 先定义标签
data_all = data_all_1 + data_all_2
label_all = label_all_1 + label_all_2
print("data_all = ",len(data_all))
print("label_all = ",len(label_all))
data_train,label_train,data_test,label_tast = data_hss.Dataset_partition(data_all,label_all)
print(data_train.shape,label_train.shape,data_test.shape,label_tast.shape)
sess = tf.InteractiveSession()
input_data = tf.placeholder(tf.float32, [None, 75])
input_label = tf.placeholder(tf.float32, [None, 2])
inputdata = tf.reshape(input_data,[-1,75,1])
conv1_7x7 = conv1d_relu(inputdata, 64, 7, 2)
pool1_3x3 = tf.layers.max_pooling1d(conv1_7x7, 3, 2, padding='same')
#深度学习的局部响应归一化LRN层，对局部神经元的活动创建竞争机制，使得其中响应比较大的值变得相对更大，并抑制其他反馈较小的神经元，增强了模型的泛化能力。
# pool1_3x3 = tf.nn.local_response_normalization(pool1_3x3) # 图像
conv2_3x3_reduce = conv1d_relu(pool1_3x3, 64, 1, 1)
conv2_3x3 = conv1d_relu(conv2_3x3_reduce, 192, 3, 1)
# conv2_3x3 = tf.nn.local_response_normalization(conv2_3x3)
pool2_3x3 = tf.layers.max_pooling1d(conv2_3x3, 3, 2, padding='same')
# 3a ###########
inception_3a_1_1 = conv1d_relu(pool2_3x3, 64, 1, 1)
inception_3a_3_3_reduce = conv1d_relu(pool2_3x3, 96, 1, 1)
inception_3a_3_3 = conv1d_relu(inception_3a_3_3_reduce, 128, 3, 1)
inception_3a_5_5_reduce = conv1d_relu(pool2_3x3, 16, 1, 1)
inception_3a_5_5 = conv1d_relu(inception_3a_5_5_reduce, 32, 5, 1)
inception_3a_pool = tf.layers.max_pooling1d(pool2_3x3, 3, 1, padding='same')
inception_3a_pool_1_1 = conv1d_relu(inception_3a_pool, 32, 1, 1)
inception_3a_output = tf.concat([inception_3a_1_1,inception_3a_3_3,inception_3a_5_5,inception_3a_pool_1_1],axis=2)
# 3b ###########
inception_3b_1_1 = conv1d_relu(inception_3a_output, 128, 1, 1)
inception_3b_3_3_reduce = conv1d_relu(inception_3a_output, 128, 1, 1)
inception_3b_3_3 = conv1d_relu(inception_3b_3_3_reduce, 192, 3, 1)
inception_3b_5_5_reduce = conv1d_relu(inception_3a_output, 32, 1, 1)
inception_3b_5_5 = conv1d_relu(inception_3b_5_5_reduce, 96, 5, 1)
inception_3b_pool = tf.layers.max_pooling1d(inception_3a_output, 3, 1, padding='same')
inception_3b_pool_1_1 = conv1d_relu(inception_3b_pool, 64, 1, 1)
inception_3b_output = tf.concat([inception_3b_1_1, inception_3b_3_3, inception_3b_5_5,inception_3b_pool_1_1],axis=2)
pool3_3_3 = tf.layers.max_pooling1d(inception_3b_output, 3, 2, padding='same')
# 4a ###########
inception_4a_1_1 = conv1d_relu(pool3_3_3, 192, 1, 1)
inception_4a_3_3_reduce = conv1d_relu(pool3_3_3, 96, 1, 1)
inception_4a_3_3 = conv1d_relu(inception_4a_3_3_reduce, 208, 3, 1)
inception_4a_5_5_reduce = conv1d_relu(pool3_3_3, 16, 1, 1)
inception_4a_5_5 = conv1d_relu(inception_4a_5_5_reduce, 48, 5, 1)
inception_4a_pool = tf.layers.max_pooling1d(pool3_3_3, 3, 1, padding='same')
inception_4a_pool_1_1 = conv1d_relu(inception_4a_pool, 64, 1, 1)
inception_4a_output = tf.concat([inception_4a_1_1, inception_4a_3_3, inception_4a_5_5, inception_4a_pool_1_1],axis=2)
# 4b ###########
inception_4b_1_1 = conv1d_relu(inception_4a_output, 160, 1, 1)
inception_4b_3_3_reduce = conv1d_relu(inception_4a_output, 112, 1, 1)
inception_4b_3_3 = conv1d_relu(inception_4b_3_3_reduce, 224, 3, 1)
inception_4b_5_5_reduce = conv1d_relu(inception_4a_output, 24, 1, 1)
inception_4b_5_5 = conv1d_relu(inception_4b_5_5_reduce, 64, 5, 1)
inception_4b_pool = tf.layers.max_pooling1d(inception_4a_output, 3, 1, padding='same')
inception_4b_pool_1_1 = conv1d_relu(inception_4b_pool, 64, 1, 1)
inception_4b_output = tf.concat([inception_4b_1_1, inception_4b_3_3, inception_4b_5_5,inception_4b_pool_1_1],axis=2)
# 4c ###########
inception_4c_1_1 = conv1d_relu(inception_4b_output, 128, 1, 1)
inception_4c_3_3_reduce = conv1d_relu(inception_4b_output, 128, 1, 1)
inception_4c_3_3 = conv1d_relu(inception_4c_3_3_reduce, 256, 3, 1)
inception_4c_5_5_reduce = conv1d_relu(inception_4b_output, 24, 1, 1)
inception_4c_5_5 = conv1d_relu(inception_4c_5_5_reduce, 64, 5, 1)
inception_4c_pool = tf.layers.max_pooling1d(inception_4b_output, 3, 1, padding='same')
inception_4c_pool_1_1 = conv1d_relu(inception_4c_pool, 64, 1, 1)
inception_4c_output = tf.concat([inception_4c_1_1, inception_4c_3_3, inception_4c_5_5, inception_4c_pool_1_1],axis=2)
# 4d ###########
inception_4d_1_1 = conv1d_relu(inception_4c_output, 112, 1, 1)
inception_4d_3_3_reduce = conv1d_relu(inception_4c_output, 144, 1, 1)
inception_4d_3_3 = conv1d_relu(inception_4d_3_3_reduce, 288, 3, 1)
inception_4d_5_5_reduce = conv1d_relu(inception_4c_output, 32, 1, 1)
inception_4d_5_5 = conv1d_relu(inception_4d_5_5_reduce, 64, 5, 1)
inception_4d_pool = tf.layers.max_pooling1d(inception_4c_output, 3, 1, padding='same')
inception_4d_pool_1_1 = conv1d_relu(inception_4d_pool, 64, 1, 1)
inception_4d_output = tf.concat([inception_4d_1_1, inception_4d_3_3, inception_4d_5_5, inception_4d_pool_1_1],axis=2)
# 4e ###########
inception_4e_1_1 = conv1d_relu(inception_4d_output, 256, 1, 1)
inception_4e_3_3_reduce = conv1d_relu(inception_4d_output, 160, 1, 1)
inception_4e_3_3 = conv1d_relu(inception_4e_3_3_reduce, 320, 3, 1)
inception_4e_5_5_reduce = conv1d_relu(inception_4d_output, 32, 1, 1)
inception_4e_5_5 = conv1d_relu(inception_4e_5_5_reduce, 128, 5, 1)
inception_4e_pool = tf.layers.max_pooling1d(inception_4d_output, 3, 1, padding='same')
# 4e ###########
inception_4e_pool_1_1 = conv1d_relu(inception_4e_pool, 128, 1, 1)
inception_4e_output = tf.concat([inception_4e_1_1, inception_4e_3_3, inception_4e_5_5,inception_4e_pool_1_1],axis=2)
pool4_3_3 = tf.layers.max_pooling1d(inception_4e_output, 3, 2, padding='same')
# 5a ###########
inception_5a_1_1 = conv1d_relu(pool4_3_3, 256, 1, 1)
inception_5a_3_3_reduce = conv1d_relu(pool4_3_3, 160, 1, 1)
inception_5a_3_3 = conv1d_relu(inception_5a_3_3_reduce, 320, 3, 1)
inception_5a_5_5_reduce = conv1d_relu(pool4_3_3, 32, 1, 1)
inception_5a_5_5 = conv1d_relu(inception_5a_5_5_reduce, 128, 5, 1)
inception_5a_pool = tf.layers.max_pooling1d(pool4_3_3, 3, 1, padding='same')
inception_5a_pool_1_1 = conv1d_relu(inception_5a_pool, 128, 1, 1)
inception_5a_output = tf.concat([inception_5a_1_1, inception_5a_3_3, inception_5a_5_5, inception_5a_pool_1_1],axis=2)
# 5b ###########
inception_5b_1_1 = conv1d_relu(inception_5a_output, 384, 1, 1)
inception_5b_3_3_reduce = conv1d_relu(inception_5a_output, 192, 1, 1)
inception_5b_3_3 = conv1d_relu(inception_5b_3_3_reduce, 384, 3, 1)
inception_5b_5_5_reduce = conv1d_relu(inception_5a_output, 48, 1, 1)
inception_5b_5_5 = conv1d_relu(inception_5b_5_5_reduce, 128, 5, 1)
inception_5b_pool = tf.layers.max_pooling1d(inception_5a_output, 3, 1, padding='same')
inception_5b_pool_1_1 = conv1d_relu(inception_5b_pool, 128, 1, 1)
inception_5b_output = tf.concat([inception_5b_1_1, inception_5b_3_3, inception_5b_5_5, inception_5b_pool_1_1],axis=2)
##########
keep_prob = tf.placeholder(tf.float32)
pool5_7_7 = tf.layers.average_pooling1d(inception_3a_output, 3, 1, padding='valid') # inception_5b_output
pool5_7_7 = tf.layers.dropout(pool5_7_7, keep_prob)
print("pool5_7_7 = ",pool5_7_7.shape)
# 铺平图像数据
pool5_7_7_flat = tf.layers.Flatten()(pool5_7_7)
print("pool5_7_7_flat = ",pool5_7_7_flat.shape)
#全连接层计算
y_conv = tf.layers.dense(pool5_7_7_flat,2) #,activation = tf.nn.softmax
# print("y_conv = ",y_conv.shape)
# 计算交叉熵损失
cross_entropy = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=y_conv, labels=input_label))
# 创建优化器,通知Tensorflow在训练时要更新均值和方差的分布
with tf.control_dependencies(tf.get_collection(tf.GraphKeys.UPDATE_OPS)):
train_step = tf.train.AdamOptimizer(1e-2).minimize(cross_entropy)
# 创建计算准确度的操作
correct_prediction = tf.equal(tf.argmax(y_conv, 1), tf.argmax(input_label, 1))
accuracy_rate = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
########开始训练过程########
# 初始化所有变量、参数
tf.global_variables_initializer().run()
m_saver = tf.train.Saver(var_list=tf.global_variables(),max_to_keep = 15) #声明tf.train.Saver类用于保存模型
########### 训练（设置训练时Dropout的kepp_prob比率为0.5。mini-batch为50，进行2000次迭代训练，参与训练样本5万）
model_doc = r'model' # 存储模型的文件夹
if not os.path.exists(model_doc): # 模型存储文件夹
os.makedirs(model_doc)
else:
shutil.rmtree(model_doc) #先强制删除文件夹
os.makedirs(model_doc)
train_accuracy_all = [] #保存所有准确度
max_acc = 0
f = open('model/acc.txt','w')
for i in range(10000):
batch_data_train,batch_label_train = data_hss.batch(data_train,label_train,batch_size = 200)
# print("batch_data_train = ",batch_data_train,batch_data_train.shape)
# print("batch_label_train = ",batch_label_train,batch_label_train.shape)
# 循环次数是100的倍数的时候，打印
if i%1 == 0:
train_accuracy = accuracy_rate.eval(feed_dict={input_data:batch_data_train,input_label: batch_label_train,
keep_prob: 0.5})
print ("-->step %d, training accuracy %.4f ,max_acc %.4f"%(i, train_accuracy,max_acc))
print("cross_entropy = ",sess.run(cross_entropy,feed_dict={input_data:batch_data_train,input_label: batch_label_train,
keep_prob: 0.5}))
f.write(str(i)+', train_accuracy: '+str(train_accuracy)+' '+str(max_acc) +'\n')
# #保存最近的5次模型
# m_saver.save(sess, './model/model.ckpt', global_step=i)
#保存准确度最高的5次模型
if train_accuracy >= max_acc :
max_acc = train_accuracy
if train_accuracy >= max_acc - 0.04:
m_saver.save(sess, './model/model.ckpt', global_step=i)
train_accuracy_all.append(train_accuracy)
# 执行训练模型
train_step_,loss = sess.run([train_step,cross_entropy], feed_dict={input_data:batch_data_train,
input_label: batch_label_train,
keep_prob: 0.5})
# 打印测试集正确率 ，全部训练完成之后，在最终测试集上进行全面测试，得到整体的分类准确率
# print(train_accuracy_all)
if max_acc >= 0.9 and train_accuracy < 0.5: #训练准确度过了最高点并下降，就停止训练
break
f.close()
#######模型读回及预测
model_file=tf.train.latest_checkpoint('model/')
m_saver.restore(sess,model_file)
#写文件
# output1 = "伪差.txt"
# if os.path.exists(output1): os.remove(output1)
# file_name1 = open(output1, 'a+')
# output2 = "非伪差.txt"
# if os.path.exists(output2): os.remove(output2)
# file_name2 = open(output2, 'a+')
# output3 = "R_left2right.txt"
# if os.path.exists(output3): os.remove(output3)
# file_name3 = open(output3, 'a+')
# output4 = "R_left2other.txt"
# if os.path.exists(output4): os.remove(output4)
# file_name4 = open(output4, 'a+')
a = 0 # 预测对的个数
TP = 0 # 预测正确的病症个数
FN_TP = 0 # 原标签中有病症的个数
TN = 0 # 预测正确的非病症个数
TN_FP = 0 # 原标签中正常的个数
sensibility = 0 # 敏感性
specificity = 0 # 特异性
single_data = np.empty([1,75])
single_label = np.zeros([1,2],"int")
for i in range(0,len(data_test)):
single_data[0] = data_test[i]
single_label = label_tast[i]
# print("single_data = ",single_data)
# print("single_label = ",single_label)
output = sess.run(y_conv , feed_dict = {input_data:single_data, keep_prob: 1.0})
# print("output = ",output)
output = sess.run(tf.nn.softmax(output))
output = np.round(output) #np.round(output) #取最接近的整数
print(i,"/",len(data_test)-1," output = ",output,"single_label = ",single_label)
if single_label[0] == output[0][0] and single_label[1] == output[0][1] :
a +=1
if single_label[0] == output[0][0] and output[0][0] == 1: #敏感性
TP += 1
if single_label[0] == 1 :
FN_TP += 1
if single_label[1] == output[0][1] and output[0][1] == 1: #特异性
TN += 1
if single_label[1] == 1 :
TN_FP += 1
# if output[0][0] == 1: # 伪差
# file_name1.write(str(data_test[i]) + '\n')
# if output[0][1] == 1: # 非伪差
# file_name2.write(str(data_test[i]) + '\n')
# if single_label[0] == 1 and output[0][0] == 0: # 伪差2非伪差
# file_name3.write(str(data_test[i]) + '\n')
# if single_label[1] == 1 and output[0][0] == 1: # 非伪差2伪差
# file_name4.write(str(data_test[i]) + '\n')
print("len(data_test) = ",len(data_test),"a =",a)
print("sensibility = ",TP/FN_TP,"specificity =",TN/TN_FP)
# file_name1.close()
# file_name2.close()
# file_name3.close()
# file_name4.close()
end = time.time()
print("程序运行时间：",end - start)
inception_v1网络
最后编辑于 ：
©著作权归作者所有,转载或内容合作请联系作者
【社区内容提示】社区部分内容疑似由AI辅助生成，浏览时请结合常识与多方信息审慎甄别。
平台声明：文章内容（如有图片或视频亦包括在内）由作者上传并发布，文章内容仅代表作者本人观点，简书系信息发布平台，仅提供信息存储服务。
【社区内容提示】社区部分内容疑似由AI辅助生成，浏览时请结合常识与多方信息审慎甄别。
平台声明：文章内容（如有图片或视频亦包括在内）由作者上传并发布，文章内容仅代表作者本人观点，简书系信息发布平台，仅提供信息存储服务。
相关阅读更多精彩内容
- 原文链接：blog.csdn.net/langb2014/article/details/52787095 来自c...
- 论文分享 萝卜兔编辑整理 卷积神经网络广泛应用在各种任务，比如图像分类、目标检测等，性能也越来越好，但都趋向于使...
- SiameseFC: Luca Bertinetto, Jack Valmadre, João F. Henriq...
