使用深度学习的VGG-16卷积神经网络训练CIFAR10数据集代码实现

CIFAR10数据集

• 包含50000张训练图片
• 包含10000张测试图片
• 共分10个类别

代码

import tensorflow as tf
import os
import pickle
import numpy as np

cifar_dir = './cifar-10-batches-py'
print(os.listdir(cifar_dir))

# 构建文件路径
train_filenames = [os.path.join(cifar_dir,'data_batch_%d'%i)for i in range(1,6)]
test_filenames = [os.path.join(cifar_dir,'test_batch')]


# 数据载入函数
def load_data(filename):
    with open(filename,'rb') as f:
        data = pickle.load(f,encoding='bytes')
        return data[b'data'],data[b'labels']


# 数据处理类
class CifarData:
    def __init__(self,filenames,need_shuffle):
        all_data = []
        all_labels = []
        for filename in filenames:
            data, labels = load_data(filename)
            all_data.append(data)
            all_labels.append(labels)

        self._data = np.vstack(all_data) / 127.5 - 1
        self._labels = np.hstack(all_labels)
        self._num_examples = self._data.shape[0]
        self._index = 0
        self._need_shuffle = need_shuffle
        if self._need_shuffle:
            self.shuffle_data()

    def shuffle_data(self):
        o = np.random.p