首页 > 程序开发 > 软件开发 > 其他 >

tensorflow1.1/构建深度卷积神经网络识别物体识别

2017-07-21

tensorflow1 1 构建深度卷积神经网络识别物体识别。

环境:tensorflow1.1,python3

#coding:utf-8
"""
python3
tensorflow 1.1
matplotlib 2.02
cifar文件包含训练集50000张,测试集合10000张
数据集为matlab类型
data_batch_1训练集样本数据10000张
data_batch_2训练集样本数据10000张
data_batch_3训练集样本数据10000张
data_batch_4训练集样本数据10000张
data_batch_5训练集样本数据10000张
"""
import tensorflow as tf
import scipy.io as sio
import numpy as np
import matplotlib.pyplot as plt

learning_rate = 0.0001
batch_size = 128
#读取数据
def read_data(filename):
    with open(filename,'rb') as f:
        dict = sio.loadmat(f)
        #图片维度要变换
        return np.array(dict['data']).reshape(10000,3,32,32).transpose(0,2,3,1),np.array(dict['labels']).reshape(10000,1)


#将图片转为灰度图
def to4d(img):
    return img.reshape(img.shape[0],32,32,3).astype(np.float32)/255

#定义将label转为one_hot
def label_to_one_hot(labels_dense, num_classes=10):
    num_labels = labels_dense.shape[0]
    index_offset = np.arange(num_labels) * num_classes
    labels_one_hot = np.zeros((num_labels, num_classes))
    labels_one_hot.flat[index_offset + labels_dense.ravel()] = 1
    return labels_one_hot

train_total_data,train_total_labels = read_data('/home/tuoxin/mytensorflow/my_project/project3/cifar-10-batches-mat/data_batch_1.mat')

#将数据集合并
for batch_num in range(2,6):
    train_data,train_labels = read_data('/home/tuoxin/mytensorflow/my_project/project3/cifar-10-batches-mat/data_batch_'+str(batch_num)+'.mat')
    train_total_data = np.vstack((train_total_data,train_data))
    train_total_labels = np.vstack((train_total_labels,train_labels))

#train_data,train_labels = read_data('/home/tuoxin/mytensorflow/my_project/project3/cifar-10-batches-py/data_batch_0')
train_data = to4d(train_total_data)
train_labels = label_to_one_hot(train_total_labels,10)
test_data,test_labels = read_data('/home/tuoxin/mytensorflow/my_project/project3/cifar-10-batches-mat/test_batch.mat')
test_data = to4d(test_data)
test_labels = label_to_one_hot(test_labels,10)

#查看图片
plt.imshow(train_data[0])
plt.title('the object is:',fontdict={'size':'16','color':'c'})
plt.show()

#定义输入形状
xs = tf.placeholder(tf.float32,[None,32,32,3])
ys = tf.placeholder(tf.float32,[None,10])
keep_prob1 = tf.placeholder(tf.float32)
keep_prob2 = tf.placeholder(tf.float32)

#构建神经网络
conv1 = tf.layers.conv2d(inputs=xs,filters=32,kernel_size=3,strides=1,padding='same',activation=tf.nn.relu)
#[None,28,28,3] --->>>[None,32,32,32]
conv2 = tf.layers.conv2d(conv1,filters=32,kernel_size=3,strides=1,padding='valid',activation=tf.nn.relu)
#[None,32,32,32] --->>>[None,30,30,32]
pool1 = tf.layers.max_pooling2d(conv2,pool_size=2,strides=2)
#[None,30,30,32] --->>>[None,15,15,32]
dropout1 = tf.nn.dropout(pool1,keep_prob1)
conv3 = tf.layers.conv2d(dropout1,filters=64,kernel_size=3,strides=1,padding='same',activation=tf.nn.relu)
#[None,15,15,32] --->>>[None,15,15,64]
conv4 = tf.layers.conv2d(conv3,filters=64,kernel_size=3,strides=1,padding='valid',activation=tf.nn.relu)
#[None,15,15,64] --->>>[None,13,13,64]
pool2 = tf.layers.max_pooling2d(conv4,pool_size=2,strides=2)
#[None,13,13,64] --->>>[None,6,6,64]
dropout2 = tf.layers.dropout(pool2,keep_prob1)
flat = tf.reshape(dropout2,[-1,6*6*64])
dense = tf.layers.dense(flat,512,tf.nn.relu)
dropout3 = tf.nn.dropout(dense,keep_prob2)
output = tf.layers.dense(dropout3,10)

#计算loss
loss = tf.losses.softmax_cross_entropy(onehot_labels=ys,logits=output)
train = tf.train.AdamOptimizer(learning_rate).minimize(loss)
#计算accuracy,返回两个参数,acc和update_op
accuracy = tf.metrics.accuracy(labels=tf.argmax(ys,axis=1),predictions=tf.argmax(output,axis=1))[1]

with tf.Session() as sess:
    init = tf.group(tf.global_variables_initializer(),tf.local_variables_initializer())
    sess.run(init)
    for step in range(1000):
        i = 0
        while i < len(train_data):
            #按批进行训练
            start = i
            end = i + batch_size
            batch_x = np.array(train_data[start:end])
            batch_y = np.array(train_labels[start:end])
            _,c = sess.run([train,loss],feed_dict={xs:batch_x,ys:batch_y,keep_prob1:0.75,keep_prob2:0.5})
            i += batch_size

        if step % 1 ==0:
            acc = sess.run(accuracy,feed_dict={xs:test_data,ys:test_labels,keep_prob1:0.75,keep_prob2:0.5})
            print(&#39;= = = = = => > > > > >&#39;,&#39;step:&#39;,step,&#39;loss: %.4f&#39; %c,&#39;accuracy:%.2f&#39; %acc)

结果:

训练速度太慢了,精度达到70%时手动终止了
这里写图片描述

这里写图片描述

相关文章
最新文章
热点推荐