06-TensorFlow 自制数据集

1.数据文件介绍

数据集下载:https://download.csdn.net/download/qq_41865229/85254826
训练集60000张数字图片, 测试集10000张图片.

都是黑底白字的灰度图,每张图有28行28列个像素点, 命名格式为 序号_数值


两个文本文件中, 每行分别是图片名, 标签(对应的数值)

路径参数:

train_path = './mnist_image_label/mnist_train_jpg_60000/'
train_txt = './mnist_image_label/mnist_train_jpg_60000.txt'
x_train_savepath = './mnist_image_label/mnist_x_train.npy'
y_train_savepath = './mnist_image_label/mnist_y_train.npy'

test_path = './mnist_image_label/mnist_test_jpg_10000/'
test_txt = './mnist_image_label/mnist_test_jpg_10000.txt'
x_test_savepath = './mnist_image_label/mnist_x_test.npy'
y_test_savepath = './mnist_image_label/mnist_y_test.npy'

2.生成数据集

我们就是要把图片转化为离散的数据, 也就是数组形式(特征值数组和标签数组)

#生成数据集
def generateds(path, txt):
    f = open(txt, 'r')  # 以只读形式打开txt文件
    contents = f.readlines()  # 读取文件中所有行
    f.close()  # 关闭txt文件
    x, y_ = [], []  # 建立空列表
    for content in contents:  # 逐行取出
        value = content.split()  # 以空格分开，图片路径为value[0] , 标签为value[1] , 存入列表
        img_path = path + value[0]  # 拼出图片路径和文件名
        img = Image.open(img_path)  # 读入图片
        img = np.array(img.convert('L'))  # 图片变为8位宽灰度值的np.array格式
        img = img / 255.  # 数据归一化 （实现预处理）
        x.append(img)  # 归一化后的数据，贴到列表x
        y_.append(value[1])  # 标签贴到列表y_
        print('loading : ' + content)  # 打印状态提示

    x = np.array(x)  # 变为np.array格式
    y_ = np.array(y_)  # 变为np.array格式
    y_ = y_.astype(np.int64)  # 变为64位整型
    print("x的类型", x.shape)
    print("y_的类型", y_.shape)
    return x, y_  # 返回输入特征x，返回标签y_

输入训练集文件的路径生成数据集, 得到60000个28*28的特征数值的矩阵数组和 60000个标签数值的一维数组

3.存储数据集

如果每次都要读入图片文件, 然后再生成数据集数组, 就很费时间; 我们可以把生成的数据集数组存储为文件, 下次要使用就可以直接读入数据集数组文件了,比较快速.

def load_data():
    #如果数据集数组文件存在, 就直接载入
    if os.path.exists(x_train_savepath) and os.path.exists(y_train_savepath) and os.path.exists(
            x_test_savepath) and os.path.exists(y_test_savepath):
        print('-------------Load Datasets-----------------')
        x_train_save = np.load(x_train_savepath)
        y_train = np.load(y_train_savepath)
        x_test_save = np.load(x_test_savepath)
        y_test = np.load(y_test_savepath)
        x_train = np.reshape(x_train_save, (len(x_train_save), 28, 28))
        x_test = np.reshape(x_test_save, (len(x_test_save), 28, 28))
    #否则读取图片文件生成数据集数组, 并存储为数组文件
    else:
        print('-------------Generate Datasets-----------------')
        print("生成训练集")
        x_train, y_train = generateds(train_path, train_txt)
        print("生成测试集")
        x_test, y_test = generateds(test_path, test_txt)

        print('-------------Save Datasets-----------------')
        x_train_save = np.reshape(x_train, (len(x_train), -1))
        x_test_save = np.reshape(x_test, (len(x_test), -1))
        np.save(x_train_savepath, x_train_save)
        np.save(y_train_savepath, y_train)
        np.save(x_test_savepath, x_test_save)
        np.save(y_test_savepath, y_test)
    return x_train, y_train, x_test, y_test

1.初始未生成数据集数组文件

2.当有已生成的数据集数组文件时, 就可以直接载入

4.使用数据集

接下来我们就可以使用数据集进行神经网络的训练了, 下面是项目完整代码

image_preprocess.py

from PIL import Image
import numpy as np
import os

train_path = './mnist_image_label/mnist_train_jpg_60000/'
train_txt = './mnist_image_label/mnist_train_jpg_60000.txt'
x_train_savepath = './mnist_image_label/mnist_x_train.npy'
y_train_savepath = './mnist_image_label/mnist_y_train.npy'

test_path = './mnist_image_label/mnist_test_jpg_10000/'
test_txt = './mnist_image_label/mnist_test_jpg_10000.txt'
x_test_savepath = './mnist_image_label/mnist_x_test.npy'
y_test_savepath = './mnist_image_label/mnist_y_test.npy'

#生成数据集
def generateds(path, txt):
    f = open(txt, 'r')  # 以只读形式打开txt文件
    contents = f.readlines()  # 读取文件中所有行
    f.close()  # 关闭txt文件
    x, y_ = [], []  # 建立空列表
    for content in contents:  # 逐行取出
        value = content.split()  # 以空格分开，图片路径为value[0] , 标签为value[1] , 存入列表
        img_path = path + value[0]  # 拼出图片路径和文件名
        img = Image.open(img_path)  # 读入图片
        img = np.array(img.convert('L'))  # 图片变为8位宽灰度值的np.array格式
        img = img / 255.  # 数据归一化 （实现预处理）
        x.append(img)  # 归一化后的数据，贴到列表x
        y_.append(value[1])  # 标签贴到列表y_
        #print('loading : ' + content)  # 打印状态提示

    x = np.array(x)  # 变为np.array格式
    y_ = np.array(y_)  # 变为np.array格式
    y_ = y_.astype(np.int64)  # 变为64位整型
    print("x的类型", x.shape)
    print("y_的类型", y_.shape)
    return x, y_  # 返回输入特征x，返回标签y_

#读入数据集
def load_data():
    #如果数据集数组文件存在, 就直接载入
    if os.path.exists(x_train_savepath) and os.path.exists(y_train_savepath) and os.path.exists(
            x_test_savepath) and os.path.exists(y_test_savepath):
        print('-------------Load Datasets-----------------')
        x_train_save = np.load(x_train_savepath)
        y_train = np.load(y_train_savepath)
        x_test_save = np.load(x_test_savepath)
        y_test = np.load(y_test_savepath)
        x_train = np.reshape(x_train_save, (len(x_train_save), 28, 28))
        x_test = np.reshape(x_test_save, (len(x_test_save), 28, 28))
    #否则读取图片文件生成数据集数组, 并存储为数组文件
    else:
        print('-------------Generate Datasets-----------------')
        print("生成训练集")
        x_train, y_train = generateds(train_path, train_txt)
        print("生成测试集")
        x_test, y_test = generateds(test_path, test_txt)

        print('-------------Save Datasets-----------------')
        x_train_save = np.reshape(x_train, (len(x_train), -1))
        x_test_save = np.reshape(x_test, (len(x_test), -1))
        np.save(x_train_savepath, x_train_save)
        np.save(y_train_savepath, y_train)
        np.save(x_test_savepath, x_test_save)
        np.save(y_test_savepath, y_test)
    return x_train, y_train, x_test, y_test

image_identification_train.py

import image_preprocess
import tensorflow as tf
#载入数据集
x_train, y_train, x_test, y_test = image_preprocess.load_data()


model = tf.keras.models.Sequential([
    tf.keras.layers.Flatten(),
    tf.keras.layers.Dense(128, activation='relu'),
    tf.keras.layers.Dense(10, activation='softmax')
])

model.compile(optimizer='adam',
              loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=False),
              metrics=['sparse_categorical_accuracy'])

model.fit(x_train, y_train, batch_size=32, epochs=5, validation_data=(x_test, y_test), validation_freq=1)
model.summary()

运行结果

5.数据增强（增大数据量）

数据增强(Data Augmentation)：是指对图片进行随机的旋转、翻转、裁剪、随机设置图片的亮度和对比度以及对数据进行标准化(数据的均值为0，方差为1)。通过这些操作，我们可以获得更多的图片样本，原来的一张图片可以变为多张图片，扩大了样本容量，对于提高模型的准确率和提升模型的泛化能力非常有帮助，在进行数据增强的同时也会需要消耗大量的系统资源。

引入数据增强操作后的代码
image_identification_train.py

import image_preprocess
import tensorflow as tf
from tensorflow.keras.preprocessing.image import ImageDataGenerator

#载入数据集
x_train, y_train, x_test, y_test = image_preprocess.load_data()


x_train = x_train.reshape(x_train.shape[0], 28, 28, 1)  # 给数据增加一个维度,从(60000, 28, 28)reshape为(60000, 28, 28, 1)

image_gen_train = ImageDataGenerator(
    rescale=1. / 1.,  # 如为图像，分母为255时，可归至0～1
    rotation_range=45,  # 随机45度旋转
    width_shift_range=.15,  # 宽度偏移
    height_shift_range=.15,  # 高度偏移
    horizontal_flip=False,  # 水平翻转
    zoom_range=0.5  # 将图像随机缩放阈量50％
)
image_gen_train.fit(x_train) #把x_train送入数据增强操作

model = tf.keras.models.Sequential([
    tf.keras.layers.Flatten(),
    tf.keras.layers.Dense(128, activation='relu'),
    tf.keras.layers.Dense(10, activation='softmax')
])

model.compile(optimizer='adam',
              loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=False),
              metrics=['sparse_categorical_accuracy'])

#以flow形式, 按照batch打包后送入
model.fit(image_gen_train.flow(x_train, y_train, batch_size=32), epochs=5, validation_data=(x_test, y_test),
          validation_freq=1)
model.summary()

运行结果