待训练图片
![](https://i-blog.csdnimg.cn/blog_migrate/ef3533ff3f4884966189530c466c93d9.jpeg)
![](https://i-blog.csdnimg.cn/blog_migrate/a1ec7ddaf02d7991405b6b1d38fb4734.jpeg)
![](https://i-blog.csdnimg.cn/blog_migrate/11541b7584ba6b6e46f55efa491ee6f3.jpeg)
![](https://i-blog.csdnimg.cn/blog_migrate/39b0440ff96bd9298c8483edfbfb9dd9.jpeg)
![](https://i-blog.csdnimg.cn/blog_migrate/5fb7086fa33c858c6ce88c7cbd9a5b25.jpeg)
![](https://i-blog.csdnimg.cn/blog_migrate/c76d294aa8bbc4d995e6cd5d6312b2ee.jpeg)
![](https://i-blog.csdnimg.cn/blog_migrate/00adf3f8c2a69e155d24630213c6e738.jpeg)
![](https://i-blog.csdnimg.cn/blog_migrate/3249dab3f991c42111708e23e0f90bf3.jpeg)
![](https://i-blog.csdnimg.cn/blog_migrate/58c094bb639e8d5e1c8973c590547af3.jpeg)
![](https://i-blog.csdnimg.cn/blog_migrate/58ee3c7d0f6a06a3867b53b43bb6c955.jpeg)
提取图片特征,存成npy文件
import multiprocessing
import os, time, random
import numpy as np
import cv2
import os
import sys
from time import ctime
import tensorflow as tf
image_dir = r"D:/sxl/处理图片/汉字分类/train10/" #图像文件夹路径
data_type = 'test'
save_path = r'E:/sxl_Programs/Python/CNN/npy/' #存储路径
data_name = 'Img10' #npy文件名
char_set = np.array(os.listdir(image_dir)) #文件夹名称列表
np.save(save_path+'ImgShuZi10.npy',char_set) #文件夹名称列表
char_set_n = len(char_set) #文件夹列表长度
read_process_n = 1 #进程数
repate_n = 4 #随机移动次数
data_size = 1000000 #1个npy大小
shuffled = True #是否打乱
#可以读取带中文路径的图
def cv_imread(file_path,type=0):
cv_img=cv2.imdecode(np.fromfile(file_path,dtype=np.uint8),-1)
# print(file_path)
# print(cv_img.shape)
# print(len(cv_img.shape))
if(type==0):
if(len(cv_img.shape)==3):
cv_img = cv2.cvtColor(cv_img, cv2.COLOR_BGR2GRAY)
return cv_img
#多个数组按同一规则打乱数据
def ShuffledData(features,labels):
'''
@description:随机打乱数据与标签,但保持数据与标签一一对应
@author:RenHui
'''
permutation = np.random.permutation(features.shape[0])
shuffled_features = features[permutation,:] #多维
shuffled_labels = labels[permutation] #1维
return shuffled_features,shuffled_labels
#函数功能:简单网格
#函数要求:1.无关图像大小;2.输入图像默认为灰度图;3.参数只有输入图像
#返回数据:1x64*64维特征
def GetFeature(image):
#图像大小归一化
image = cv2.resize(image,(64,64))
img_h = image.shape[0]
img_w = image.shape[1]
#定义特征向量
feature = np.zeros(img_h*img_w,dtype=np.float32)
for h in range(img_h):
for w in range(img_w):
feature[h*img_h+w] = image[h,w]
return feature
# 写数据进程执行的代码:
def read_image_to_queue(queue):
print('Process to write: %s' % os.getpid())
for j,dirname in enumerate(char_set): # dirname 是文件夹名称
label = np.where(char_set==dirname)[0][0] #文件夹名称对应的下标序号
print('序号:'+str(j),'读 '+dirname+' 文件夹...时间:',ctime() )
for parent,_,filenames in os.walk(os.path.join(image_dir,dirname)):
for filename in filenames:
if(filename[-4:]!='.jpg'):
continue
image = cv_imread(os.path.join(parent,filename),0)
# cv2.imshow(dirname,image)
# cv2.waitKey(0)
# cv2.imwrite(("D:/%s.jpg"%label),image)
queue.put((image,label))
a=0
for i in range(read_process_n):
queue.put((None,-1))
print('读图结束!')
return True
# 读数据进程执行的代码:
def extract_feature(queue,lock,count):
'''
@description:从队列中取出图片进行特征提取
@queue:先进先出队列
lock:锁,在计数时上锁,防止冲突
count:计数
'''
print('Process %s start reading...' % os.getpid())
global data_n
features = [] #存放提取到的特征
labels = [] #存放标签
flag = True #标志着进程是否结束
while flag:
image,label = queue.get() #从队列中获取图像和标签
if len(features) >= data_size or label == -1: #特征数组的长度大于指定长度,则开始存储
array_features = np.array(features) #转换成数组
array_labels = np.array(labels)
array_features,array_labels = ShuffledData(array_features,array_labels) # 打乱数据
array_features, array_labels = ShuffledData(array_features, array_labels) # 打乱数据
lock.acquire() # 锁开始
# 拆分数据为训练集,测试集
split_x = int(array_features.shape[0] * 0.8)
train_data, test_data = np.split(array_features, [split_x], axis=0) # 拆分特征数据集
train_labels, test_labels = np.split(array_labels, [split_x], axis=0) # 拆分标签数据集
count.value += 1 #下标计数加1
str_features_name_train = data_name+'_features_train_'+str(count.value)+'.npy'
str_labels_name_train = data_name+'_labels_train_'+str(count.value)+'.npy'
str_features_name_test = data_name+'_features_test_'+str(count.value)+'.npy'
str_labels_name_test = data_name+'_labels_test_'+str(count.value)+'.npy'
lock.release() # 锁释放
np.save(save_path+str_features_name_train,train_data)
np.save(save_path+str_labels_name_train,train_labels)
np.save(save_path+str_features_name_test,test_data)
np.save(save_path+str_labels_name_test,test_labels)
print(os.getpid(),'save:',str_features_name_train)
print(os.getpid(),'save:',str_labels_name_train)
print(os.getpid(),'save:',str_features_name_test)
print(os.getpid(),'save:',str_labels_name_test)
features.clear()
labels.clear()
if label == -1:
break
# 获取特征向量,传入灰度图
feature = GetFeature(image)
features.append(feature)
labels.append(label)
# # 随机移动4次
# for itime in range(repate_n):
# rMovedImage = randomMoveImage(image)
# feature = SimpleGridFeature(rMovedImage) # 简单网格
# features.append(feature)
# labels.append(label)
print('Process %s is done!' % os.getpid())
if __name__=='__main__':
time_start = time.time() # 开始计时
# 父进程创建Queue,并传给各个子进程:
image_queue = multiprocessing.Queue(maxsize=1000) #队列
lock = multiprocessing.Lock() #锁
count = multiprocessing.Value('i',0) #计数
#将图写入队列进程
write_sub_process = multiprocessing.Process(target=read_image_to_queue, args=(image_queue,))
read_sub_processes = [] #读图子线程
for i in range(read_process_n):
read_sub_processes.append(
multiprocessing.Process(target=extract_feature, args=(image_queue,lock,count))
)
# 启动子进程pw,写入:
write_sub_process.start()
# 启动子进程pr,读取:
for p in read_sub_processes:
p.start()
# 等待进程结束:
write_sub_process.join()
for p in read_sub_processes:
p.join()
time_end=time.time()
time_h=(time_end-time_start)/3600
print('用时:%.6f 小时'% time_h)
print ("读图提取特征存npy,运行结束!")
10分类模型训练
# !/usr/bin/env python
# coding: utf-8
import numpy as np
import os
import tensorflow as tf
import cv2
from tensorflow.python.framework import graph_util
MODEL_SAVE_PATH = "./model/"
MODEL_NAME = "data10_model"
n_label = 10 # 标签维度
batch_size = 1600 # 每个批次的大小
learning_rate = 0.001 # 初始学习速率时0.1
decay_rate = 0.96 # 衰减率
global_steps = 30001 # 总的迭代次数
decay_steps = 100 # 衰减次数
# 参数概要,tf.summary.scalar的作用主要是存储变量,并赋予变量名,tf.name_scope主要是给表达式命名
def variable_summaries(var):
with tf.name_scope('summaries'):
mean = tf.reduce_mean(var)
tf.summary.scalar('mean', mean) # 平均值
with tf.name_scope('stddev'):
stddev = tf.sqrt(tf.reduce_mean(tf.square(var - mean)))
tf.summary.scalar('stddev', stddev) # 标准差
tf.summary.scalar('max', tf.reduce_max(var)) # 最大值
tf.summary.scalar('min', tf.reduce_min(var)) # 最小值
tf.summary.histogram('histogram', var) # 直方图
# 初始化权值
def weight_variable(shape, name):
initial = tf.truncated_normal(shape, stddev=0.1) # 生成一个截断的正态分布
return tf.Variable(initial, name=name)
# 初始化偏置
def bias_variable(shape, name):
initial = tf.constant(0.1, shape=shape)
return tf.Variable(initial, name=name)
# 卷积层
def conv2d(x, W):
# x input tensor of shape [batch,in_height,in_width,in_channels]
# W filter/ kernel tensor of shape [filter_height,filter_width,in_channels,out_channels]
# strides[0]=strides[3]=1恒等于1,
# strides[1]代表x方向的步长,strides[2]代表y方向的步长,
# padding:"SAME","VALID"
return tf.nn.conv2d(x, W, strides=[1, 1, 1, 1], padding='SAME')
# 池化层
def max_pool_2x2(x):
# x input tensor of shape [batch,in_height,in_width,in_channels]
# ksize [1,x,y,1]
return tf.nn.max_pool(x, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')
# 从数字标签转换为数组标签 [0,0,0,...1,0,0]
def InitImagesLabels(labels_batch):
labels_batch_new = []
for id in labels_batch:
aa = np.zeros(n_label, np.float32)
aa[int(id)] = 1
labels_batch_new.append(aa)
return labels_batch_new
# 显示图像与标签
def ShowImageAndLabels(batch_xs, batch_ys):
img_h = 64
img_w = 64
img = np.ones((img_h, img_w), dtype=np.uint8)
icount=0
for batch_image in batch_xs: #转换成图像
for h in range(img_h):
for w in range(img_w):
img[h, w]=batch_image[h * img_h + w] #图像复原
sss="%d"%batch_ys[icount]
cv2.imshow(sss,img)
cv2.waitKey(0)
icount+=1
# keep_prob用来表示神经元的输出概率
with tf.name_scope('keep_prob'):
keep_prob = tf.placeholder(tf.float32, name='keep_prob')
# 命名空间
with tf.name_scope('input'):
# 定义两个placeholder
x = tf.placeholder(tf.float32, [None, 4096], name='x-input')
y = tf.placeholder(tf.float32, [None, 10], name='y-input')
with tf.name_scope('x_image'):
# 改变x的格式转为4D的向量[batch,in_height,in_width,in_channels]
x_image = tf.reshape(x, [-1, 64, 64, 1], name='x_image')
with tf.name_scope('Conv1'):
# 初始化第一个卷积层的权值和偏置
with tf.name_scope('W_conv1'):
W_conv1 = weight_variable([3, 3, 1, 32], name='W_conv1') # 5*5的采样窗口,32个卷积核从1个平面抽取特征
with tf.name_scope('b_conv1'):
b_conv1 = bias_variable([32], name='b_conv1') # 每一个卷积核一个偏置值
# 把x_image和权值向量进行卷积,再加上偏置值,然后应用于relu激活函数
with tf.name_scope('conv2d_1'):
conv2d_1 = conv2d(x_image, W_conv1) + b_conv1
with tf.name_scope('h_pool1'):
h_pool1 = max_pool_2x2(conv2d_1) # 进行max_pooling
with tf.name_scope('h_pool1_drop'):
h_pool1 = tf.nn.dropout(h_pool1, keep_prob, name='h_pool1_drop')
with tf.name_scope('Conv2'):
# 初始化第二个卷积层的权值和偏置
with tf.name_scope('W_conv2'):
W_conv2 = weight_variable([3, 3, 32, 64], name='W_conv2') # 5*5的采样窗口,64个卷积核从32个平面抽取特征
with tf.name_scope('b_conv2'):
b_conv2 = bias_variable([64], name='b_conv2') # 每一个卷积核一个偏置值
# 把h_pool1和权值向量进行卷积,再加上偏置值,
with tf.name_scope('conv2d_2'):
conv2d_2 = conv2d(h_pool1, W_conv2) + b_conv2
with tf.name_scope('h_pool2'):
h_pool2 = max_pool_2x2(conv2d_2) # 进行max_pooling
with tf.name_scope('h_pool2_drop'):
h_pool2 = tf.nn.dropout(h_pool2, keep_prob, name='h_pool2_drop')
with tf.name_scope('Conv3'):
# 初始化第三个卷积层的权值和偏置
with tf.name_scope('W_conv3'):
W_conv3 = weight_variable([3, 3, 64, 64], name='W_conv3') # 5*5的采样窗口,64个卷积核从64个平面抽取特征
with tf.name_scope('b_conv3'):
b_conv3 = bias_variable([64], name='b_conv3') # 每一个卷积核一个偏置值
# 把h_pool2和权值向量进行卷积,再加上偏置值,
with tf.name_scope('conv2d_3'):
conv2d_3 = conv2d(h_pool2, W_conv3) + b_conv3
with tf.name_scope('h_pool3'):
h_pool3 = max_pool_2x2(conv2d_3) # 进行max_pooling
with tf.name_scope('h_pool3_drop'):
h_pool3 = tf.nn.dropout(h_pool3, keep_prob, name='h_pool3_drop')
# 64*64的图片第一次卷积后还是64*64,第一次池化后变为32*32
# 第二次卷积后为32*32,第二次池化后变成了16*16
# 第三次卷积后为16*16,第二次池化后变成了8*8
# 经过上面操作后得到64张8*8的平面
with tf.name_scope('fc1'):
# 初始化第一个全连接层的权值
with tf.name_scope('W_fc1'):
W_fc1 = weight_variable([8 * 8 * 64, 1024], name='W_fc1') # 上一层有7*7*64个神经元,全连接层有1024个神经元
with tf.name_scope('b_fc1'):
b_fc1 = bias_variable([1024], name='b_fc1') # 1024个节点
# 把池化层3的输出扁平化为1维
with tf.name_scope('h_pool2_flat'):
h_pool2_flat = tf.reshape(h_pool3, [-1, 8 * 8 * 64], name='h_pool2_flat')
# 求第一个全连接层的输出
with tf.name_scope('wx_plus_b1'):
wx_plus_b1 = tf.matmul(h_pool2_flat, W_fc1) + b_fc1
with tf.name_scope('relu'):
# h_fc1 = tf.nn.relu(wx_plus_b1)
h_fc1 = tf.nn.sigmoid(wx_plus_b1)
with tf.name_scope('h_fc1_drop'):
h_fc1_drop = tf.nn.dropout(h_fc1, keep_prob, name='h_fc1_drop')
with tf.name_scope('fc2'):
# 初始化第二个全连接层
with tf.name_scope('W_fc2'):
W_fc2 = weight_variable([1024, 10], name='W_fc2')
with tf.name_scope('b_fc2'):
b_fc2 = bias_variable([10], name='b_fc2')
with tf.name_scope('wx_plus_b2'):
wx_plus_b2 = tf.matmul(h_fc1_drop, W_fc2) + b_fc2
with tf.name_scope('softmax'):
# 计算输出
# prediction = tf.nn.softmax(wx_plus_b2) # tf.nn.softmax_cross_entropy_with_logits 有softmax操作
prediction = wx_plus_b2
# 交叉熵代价函数
with tf.name_scope('cross_entropy'):
cross_entropy = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=y, logits=prediction),
name='cross_entropy')
tf.summary.scalar('cross_entropy', cross_entropy)
# 使用AdamOptimizer进行优化
with tf.name_scope('train'):
# 运行了几轮batch_size的计数器,初值给0,设为不被训练
global_step = tf.Variable(0, trainable=False)
train_step = tf.train.AdamOptimizer(learning_rate,).minimize(cross_entropy,global_step=global_step) # 断点续训这里不加global_step=global_step会出错
# 求准确率
with tf.name_scope('accuracy'):
with tf.name_scope('correct_prediction'):
# 结果存放在一个布尔列表中
correct_prediction = tf.equal(tf.argmax(prediction, 1), tf.argmax(y, 1)) # argmax返回一维张量中最大的值所在的位置
with tf.name_scope('accuracy'):
# 求准确率
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
tf.summary.scalar('accuracy', accuracy)
# 合并所有的summary
merged = tf.summary.merge_all()
saver = tf.train.Saver(max_to_keep=1)
with tf.Session() as sess:
print("启动执行...")
sess.run(tf.global_variables_initializer())
# 加入断点续训功能
ckpt = tf.train.get_checkpoint_state(MODEL_SAVE_PATH)
if ckpt and ckpt.model_checkpoint_path:
saver.restore(sess, ckpt.model_checkpoint_path)
train_writer = tf.summary.FileWriter('D:/logs/train', sess.graph)
test_writer = tf.summary.FileWriter('D:/logs/test', sess.graph)
# 读取数据
train_data_npyPath = (r"D:/SXL/npy/Img10_features_train_1.npy") # npy路径
train_labels_npyPath = (r"D:/SXL/npy/Img10_labels_train_1.npy") # npy路径
test_data_npyPath = (r"D:/SXL/npy/Img10_features_test_1.npy") # npy路径
test_labels_npyPath = (r"D:/SXL/npy/Img10_labels_test_1.npy") # npy路径
train_data = np.load(train_data_npyPath).astype(np.float32) # 加载数据
train_labels = np.load(train_labels_npyPath).astype(np.float32) # 加载数据
test_data = np.load(test_data_npyPath).astype(np.float32) # 加载数据
test_labels = np.load(test_labels_npyPath).astype(np.float32) # 加载数据
step = 0
while step < global_steps:
# 训练模型
# 记录训练集计算的参数
flag = step
if ((flag + 1) * batch_size > len(train_data)):
flag = 0
batch_xs, batch_ys = train_data[flag * batch_size:(flag + 1) * batch_size], train_labels[flag * batch_size:(flag + 1) * batch_size] # 获取一块最小数据集
batch_ys_new = InitImagesLabels(batch_ys) # 从数字标签转换为数组标签 [0,0,0,...1,0,0]
_, loss, step= sess.run([train_step, cross_entropy, global_step], feed_dict={x: batch_xs, y: batch_ys_new, keep_prob: 0.5})
summary = sess.run(merged, feed_dict={x: batch_xs, y: batch_ys_new, keep_prob: 1.0})
train_writer.add_summary(summary, step)
# 计算测试集计算的参数
flag2 = step
if ((flag2 + 1) * batch_size > len(test_data)):
flag2 = 0
batch_xs, batch_ys = test_data[flag2 * batch_size:(flag2 + 1) * batch_size], test_labels[flag2 * batch_size:(flag2 + 1) * batch_size] # 获取一块最小数据集
batch_ys_new = InitImagesLabels(batch_ys) # 从数字标签转换为数组标签 [0,0,0,...1,0,0]
summary = sess.run(merged, feed_dict={x: batch_xs, y: batch_ys_new, keep_prob: 1.0})
test_writer.add_summary(summary, step)
if step % 100 == 0:
saver.save(sess, os.path.join(MODEL_SAVE_PATH, MODEL_NAME), global_step=global_step)
print(step,":",loss)
if step % 100 == 0:
# 训练集正确率
icount_train_batchsize = int(len(train_data) / batch_size)
sum_train_acc = 0
for iNo in range(0, icount_train_batchsize):
train_batch_xs, train_batch_ys = train_data[iNo * batch_size:(iNo + 1) * batch_size], train_labels[iNo * batch_size:(iNo + 1) * batch_size] # 获取一块最小数据集
train_labels_new = InitImagesLabels(train_batch_ys) # 从数字标签转换为数组标签 [0,0,0,...1,0,0]
train_acc = sess.run(accuracy, feed_dict={x: train_batch_xs, y: train_labels_new, keep_prob: 1.0})
sum_train_acc = sum_train_acc + train_acc
sum_train_acc = sum_train_acc / icount_train_batchsize
print("Iter " + str(step) + ",Training Accuracy=" + str(sum_train_acc))
# 测试集正确率
icount_test_batchsize = int(len(test_data) / batch_size)
sum_test_acc = 0
for iNo in range(0, icount_test_batchsize):
test_batch_xs, test_batch_ys = test_data[iNo * batch_size:(iNo + 1) * batch_size], test_labels[iNo * batch_size:(iNo + 1) * batch_size] # 获取一块最小数据集
test_labels_new = InitImagesLabels(test_batch_ys) # 从数字标签转换为数组标签 [0,0,0,...1,0,0]
test_acc = sess.run(accuracy, feed_dict={x: test_batch_xs, y: test_labels_new, keep_prob: 1.0})
sum_test_acc = sum_test_acc + test_acc
sum_test_acc = sum_test_acc / icount_test_batchsize
print("Iter " + str(step) + ",Testing Accuracy=" + str(sum_test_acc))
训练结果:
![](https://i-blog.csdnimg.cn/blog_migrate/32350cdb936ab001a2dc517b3c1aa582.png)
![](https://i-blog.csdnimg.cn/blog_migrate/30a7a4d6c5d1a37841a831d75cfe99a2.png)
![](https://i-blog.csdnimg.cn/blog_migrate/1743cf9d9d1a916da34d2b85702f760f.jpeg)