服务器之家

服务器之家 > 正文

Python通过TensorFlow卷积神经网络实现猫狗识别

时间:2021-06-06 00:31     来源/作者:双斜杠少年

这份数据集来源于Kaggle,数据集有12500只猫和12500只狗。在这里简单介绍下整体思路

  1. 处理数据
  2. 设计神经网络
  3. 进行训练测试

1. 数据处理

将图片数据处理为 tf 能够识别的数据格式,并将数据设计批次。

  • 第一步get_files() 方法读取图片,然后根据图片名,添加猫狗 label,然后再将 image和label 放到 数组中,打乱顺序返回
  • 将第一步处理好的图片 和label 数组 转化为 tensorflow 能够识别的格式,然后将图片裁剪和补充进行标准化处理,分批次返回。

新建数据处理文件 ,文件名 input_data.py

?
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
import tensorflow as tf
import os
import numpy as np
def get_files(file_dir):
 cats = []
 label_cats = []
 dogs = []
 label_dogs = []
 for file in os.listdir(file_dir):
 name = file.split(sep='.')
 if 'cat' in name[0]:
 cats.append(file_dir + file)
 label_cats.append(0)
 else:
 if 'dog' in name[0]:
 dogs.append(file_dir + file)
 label_dogs.append(1)
 image_list = np.hstack((cats,dogs))
 label_list = np.hstack((label_cats,label_dogs))
 # print('There are %d cats\nThere are %d dogs' %(len(cats), len(dogs)))
 # 多个种类分别的时候需要把多个种类放在一起,打乱顺序,这里不需要
 # 把标签和图片都放倒一个 temp 中 然后打乱顺序,然后取出来
 temp = np.array([image_list,label_list])
 temp = temp.transpose()
 # 打乱顺序
 np.random.shuffle(temp)
 # 取出第一个元素作为 image 第二个元素作为 label
 image_list = list(temp[:,0])
 label_list = list(temp[:,1])
 label_list = [int(i) for i in label_list]
 return image_list,label_list
# 测试 get_files
# imgs , label = get_files('/Users/yangyibo/GitWork/pythonLean/AI/猫狗识别/testImg/')
# for i in imgs:
# print("img:",i)
# for i in label:
# print('label:',i)
# 测试 get_files end
# image_W ,image_H 指定图片大小,batch_size 每批读取的个数 ,capacity队列中 最多容纳元素的个数
def get_batch(image,label,image_W,image_H,batch_size,capacity):
 # 转换数据为 ts 能识别的格式
 image = tf.cast(image,tf.string)
 label = tf.cast(label, tf.int32)
 # 将image 和 label 放倒队列里
 input_queue = tf.train.slice_input_producer([image,label])
 label = input_queue[1]
 # 读取图片的全部信息
 image_contents = tf.read_file(input_queue[0])
 # 把图片解码,channels =3 为彩色图片, r,g ,b 黑白图片为 1 ,也可以理解为图片的厚度
 image = tf.image.decode_jpeg(image_contents,channels =3)
 # 将图片以图片中心进行裁剪或者扩充为 指定的image_W,image_H
 image = tf.image.resize_image_with_crop_or_pad(image, image_W, image_H)
 # 对数据进行标准化,标准化,就是减去它的均值,除以他的方差
 image = tf.image.per_image_standardization(image)
 # 生成批次 num_threads 有多少个线程根据电脑配置设置 capacity 队列中 最多容纳图片的个数 tf.train.shuffle_batch 打乱顺序,
 image_batch, label_batch = tf.train.batch([image, label],batch_size = batch_size, num_threads = 64, capacity = capacity)
 # 重新定义下 label_batch 的形状
 label_batch = tf.reshape(label_batch , [batch_size])
 # 转化图片
 image_batch = tf.cast(image_batch,tf.float32)
 return image_batch, label_batch
# test get_batch
# import matplotlib.pyplot as plt
# BATCH_SIZE = 2
# CAPACITY = 256
# IMG_W = 208
# IMG_H = 208
# train_dir = '/Users/yangyibo/GitWork/pythonLean/AI/猫狗识别/testImg/'
# image_list, label_list = get_files(train_dir)
# image_batch, label_batch = get_batch(image_list, label_list, IMG_W, IMG_H, BATCH_SIZE, CAPACITY)
# with tf.Session() as sess:
# i = 0
# # Coordinator 和 start_queue_runners 监控 queue 的状态,不停的入队出队
# coord = tf.train.Coordinator()
# threads = tf.train.start_queue_runners(coord=coord)
# # coord.should_stop() 返回 true 时也就是 数据读完了应该调用 coord.request_stop()
# try:
#  while not coord.should_stop() and i<1:
#   # 测试一个步
#   img, label = sess.run([image_batch, label_batch])
#   for j in np.arange(BATCH_SIZE):
#    print('label: %d' %label[j])
#    # 因为是个4D 的数据所以第一个为 索引 其他的为冒号就行了
#    plt.imshow(img[j,:,:,:])
#    plt.show()
#   i+=1
# # 队列中没有数据
# except tf.errors.OutOfRangeError:
#  print('done!')
# finally:
#  coord.request_stop()
# coord.join(threads)
 # sess.close()

2. 设计神经网络

利用卷积神经网路处理,网络结构为

?
1
2
3
4
5
6
7
# conv1 卷积层 1
# pooling1_lrn 池化层 1
# conv2 卷积层 2
# pooling2_lrn 池化层 2
# local3 全连接层 1
# local4 全连接层 2
# softmax 全连接层 3

新建神经网络文件 ,文件名 model.py

?
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
#coding=utf-8
import tensorflow as tf
def inference(images, batch_size, n_classes):
 with tf.variable_scope('conv1') as scope:
  # 卷积盒的为 3*3 的卷积盒,图片厚度是3,输出是16个featuremap
  weights = tf.get_variable('weights',
         shape=[3, 3, 3, 16],
         dtype=tf.float32,
         initializer=tf.truncated_normal_initializer(stddev=0.1, dtype=tf.float32))
  biases = tf.get_variable('biases',
         shape=[16],
         dtype=tf.float32,
         initializer=tf.constant_initializer(0.1))
  conv = tf.nn.conv2d(images, weights, strides=[1, 1, 1, 1], padding='SAME')
  pre_activation = tf.nn.bias_add(conv, biases)
  conv1 = tf.nn.relu(pre_activation, name=scope.name)
 with tf.variable_scope('pooling1_lrn') as scope:
   pool1 = tf.nn.max_pool(conv1, ksize=[1, 3, 3, 1], strides=[1, 2, 2, 1], padding='SAME', name='pooling1')
   norm1 = tf.nn.lrn(pool1, depth_radius=4, bias=1.0, alpha=0.001 / 9.0, beta=0.75, name='norm1')
 with tf.variable_scope('conv2') as scope:
    weights = tf.get_variable('weights',
           shape=[3, 3, 16, 16],
           dtype=tf.float32,
           initializer=tf.truncated_normal_initializer(stddev=0.1, dtype=tf.float32))
    biases = tf.get_variable('biases',
           shape=[16],
           dtype=tf.float32,
           initializer=tf.constant_initializer(0.1))
    conv = tf.nn.conv2d(norm1, weights, strides=[1, 1, 1, 1], padding='SAME')
    pre_activation = tf.nn.bias_add(conv, biases)
    conv2 = tf.nn.relu(pre_activation, name='conv2')
 # pool2 and norm2
 with tf.variable_scope('pooling2_lrn') as scope:
  norm2 = tf.nn.lrn(conv2, depth_radius=4, bias=1.0, alpha=0.001 / 9.0, beta=0.75, name='norm2')
  pool2 = tf.nn.max_pool(norm2, ksize=[1, 3, 3, 1], strides=[1, 1, 1, 1], padding='SAME', name='pooling2')
 with tf.variable_scope('local3') as scope:
  reshape = tf.reshape(pool2, shape=[batch_size, -1])
  dim = reshape.get_shape()[1].value
  weights = tf.get_variable('weights',
         shape=[dim, 128],
         dtype=tf.float32,
         initializer=tf.truncated_normal_initializer(stddev=0.005, dtype=tf.float32))
  biases = tf.get_variable('biases',
         shape=[128],
         dtype=tf.float32,
         initializer=tf.constant_initializer(0.1))
 local3 = tf.nn.relu(tf.matmul(reshape, weights) + biases, name=scope.name)
 # local4
 with tf.variable_scope('local4') as scope:
  weights = tf.get_variable('weights',
         shape=[128, 128],
         dtype=tf.float32,
         initializer=tf.truncated_normal_initializer(stddev=0.005, dtype=tf.float32))
  biases = tf.get_variable('biases',
         shape=[128],
         dtype=tf.float32,
         initializer=tf.constant_initializer(0.1))
  local4 = tf.nn.relu(tf.matmul(local3, weights) + biases, name='local4')
 # softmax
 with tf.variable_scope('softmax_linear') as scope:
  weights = tf.get_variable('softmax_linear',
         shape=[128, n_classes],
         dtype=tf.float32,
         initializer=tf.truncated_normal_initializer(stddev=0.005, dtype=tf.float32))
  biases = tf.get_variable('biases',
         shape=[n_classes],
         dtype=tf.float32,
         initializer=tf.constant_initializer(0.1))
  softmax_linear = tf.add(tf.matmul(local4, weights), biases, name='softmax_linear')
 return softmax_linear
def losses(logits, labels):
 with tf.variable_scope('loss') as scope:
  cross_entropy = tf.nn.sparse_softmax_cross_entropy_with_logits \
      (logits=logits, labels=labels, name='xentropy_per_example')
  loss = tf.reduce_mean(cross_entropy, name='loss')
  tf.summary.scalar(scope.name + '/loss', loss)
 return loss
def trainning(loss, learning_rate):
 with tf.name_scope('optimizer'):
  optimizer = tf.train.AdamOptimizer(learning_rate= learning_rate)
  global_step = tf.Variable(0, name='global_step', trainable=False)
  train_op = optimizer.minimize(loss, global_step= global_step)
 return train_op
def evaluation(logits, labels):
 with tf.variable_scope('accuracy') as scope:
  correct = tf.nn.in_top_k(logits, labels, 1)
  correct = tf.cast(correct, tf.float16)
  accuracy = tf.reduce_mean(correct)
  tf.summary.scalar(scope.name + '/accuracy', accuracy)
 return accuracy

3. 训练数据,并将训练的模型存储

?
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
import os
import numpy as np
import tensorflow as tf
import input_data 
import model
N_CLASSES = 2 # 2个输出神经元,[1,0] 或者 [0,1]猫和狗的概率
IMG_W = 208 # 重新定义图片的大小,图片如果过大则训练比较慢
IMG_H = 208
BATCH_SIZE = 32 #每批数据的大小
CAPACITY = 256
MAX_STEP = 15000 # 训练的步数,应当 >= 10000
learning_rate = 0.0001 # 学习率,建议刚开始的 learning_rate <= 0.0001
def run_training():
 # 数据集
 train_dir = '/Users/yangyibo/GitWork/pythonLean/AI/猫狗识别/img/' #My dir--20170727-csq
 #logs_train_dir 存放训练模型的过程的数据,在tensorboard 中查看
 logs_train_dir = '/Users/yangyibo/GitWork/pythonLean/AI/猫狗识别/saveNet/'
 # 获取图片和标签集
 train, train_label = input_data.get_files(train_dir)
 # 生成批次
 train_batch, train_label_batch = input_data.get_batch(train,
               train_label,
               IMG_W,
               IMG_H,
               BATCH_SIZE,
               CAPACITY)
 # 进入模型
 train_logits = model.inference(train_batch, BATCH_SIZE, N_CLASSES)
 # 获取 loss
 train_loss = model.losses(train_logits, train_label_batch)
 # 训练
 train_op = model.trainning(train_loss, learning_rate)
 # 获取准确率
 train__acc = model.evaluation(train_logits, train_label_batch)
 # 合并 summary
 summary_op = tf.summary.merge_all()
 sess = tf.Session()
 # 保存summary
 train_writer = tf.summary.FileWriter(logs_train_dir, sess.graph)
 saver = tf.train.Saver()
 sess.run(tf.global_variables_initializer())
 coord = tf.train.Coordinator()
 threads = tf.train.start_queue_runners(sess=sess, coord=coord)
 try:
  for step in np.arange(MAX_STEP):
   if coord.should_stop():
     break
   _, tra_loss, tra_acc = sess.run([train_op, train_loss, train__acc])
   if step % 50 == 0:
    print('Step %d, train loss = %.2f, train accuracy = %.2f%%' %(step, tra_loss, tra_acc*100.0))
    summary_str = sess.run(summary_op)
    train_writer.add_summary(summary_str, step)
   if step % 2000 == 0 or (step + 1) == MAX_STEP:
    # 每隔2000步保存一下模型,模型保存在 checkpoint_path 中
    checkpoint_path = os.path.join(logs_train_dir, 'model.ckpt')
    saver.save(sess, checkpoint_path, global_step=step)
 except tf.errors.OutOfRangeError:
  print('Done training -- epoch limit reached')
 finally:
  coord.request_stop()
 coord.join(threads)
 sess.close()
# train
run_training()

关于保存的模型怎么使用将在下一篇文章中展示。

TensorFlow 卷积神经网络之使用训练好的模型识别猫狗图片

如果需要训练数据集可以评论留下联系方式。

原文完整代码地址:

https://github.com/527515025/My-TensorFlow-tutorials/tree/master/猫狗识别

欢迎 star 欢迎提问。

总结

以上就是这篇文章的全部内容了,希望本文的内容对大家的学习或者工作具有一定的参考学习价值,谢谢大家对服务器之家的支持。如果你想了解更多相关内容请查看下面相关链接

原文链接:https://blog.csdn.net/u012373815/article/details/78768727

相关文章

热门资讯

2020微信伤感网名听哭了 让对方看到心疼的伤感网名大全
2020微信伤感网名听哭了 让对方看到心疼的伤感网名大全 2019-12-26
yue是什么意思 网络流行语yue了是什么梗
yue是什么意思 网络流行语yue了是什么梗 2020-10-11
背刺什么意思 网络词语背刺是什么梗
背刺什么意思 网络词语背刺是什么梗 2020-05-22
苹果12mini价格表官网报价 iPhone12mini全版本价格汇总
苹果12mini价格表官网报价 iPhone12mini全版本价格汇总 2020-11-13
2021德云社封箱演出完整版 2021年德云社封箱演出在线看
2021德云社封箱演出完整版 2021年德云社封箱演出在线看 2021-03-15
返回顶部