content

环境准备

Anaconda
docker

数据获取与处理

通过kaggle下载数据
通过keras下载数据
数据处理

模型训练

CNN-v1
CNN-v2

本项目使用fashion-MNIST数据集，模型采用keras方式进行训练并最终部署在Android上。完整请参考Github：fashionMNIST-on-device

环境准备

1. Anaconda

注：安装任何包请使用conda install xxx命令

2. docker

更推荐使用docker方式搭建自己的开发环境

动手学Docker：https://docs.docker.knowledge-precipitation.site/
推荐大家使用此深度学习docker镜像：deepo，此镜像已预装深度学习基本开发环境

>>> import tensorflow
>>> import sonnet
>>> import torch
>>> import keras
>>> import mxnet
>>> import cntk
>>> import chainer
>>> import theano
>>> import lasagne
>>> import caffe
>>> import caffe2

数据获取与处理

Fashion-MNIST是一个替代MNIST手写数字集的图像数据集。它是由Zalando（一家德国的时尚科技公司）旗下的研究部门提供。其涵盖了来自10种类别的共7万个不同商品的正面图片。Fashion-MNIST的大小、格式和训练集/测试集划分与原始的MNIST完全一致。60000/10000的训练测试数据划分，28x28的灰度图片。下载地址

1. 通过kaggle下载数据

kaggle链接

2. 通过keras下载数据

from keras.datasets import fashion_mnist

(train_images, train_labels), (test_images, test_labels) = fashion_mnist.load_data()

3. 数据处理

# 首先看一下数据的形状
print(train_images.shape)
print(test_images.shape)

#输出结果
(60000, 28, 28)
(10000, 28, 28)

可以看到训练数据是60000张28*28的图片，测试数据是10000张28*28的图片。

我们来看一下图片上都是什么数据：

import matplotlib.pyplot as plt

plt.imshow(train_images[0])
plt.savefig("train_images_0.png")
plt.show()

显示的结果：

之后将数据做reshape使数据适合训练，并将数据缩放到0-1之间。

train_images = train_images.reshape(train_images.shape[0], 28, 28, 1)
test_images = test_images.reshape(test_images.shape[0], 28, 28, 1)
train_images = train_images.astype('float32') / 255
test_images = test_images.astype('float32') / 255

对标签做one-hot编码：

train_labels = to_categorical(train_labels, 10)
test_labels = to_categorical(test_labels, 10)

将所有操作整合为一个函数：

def load_data():
    (train_images, train_labels), (test_images, test_labels) = fashion_mnist.load_data()
    train_images = train_images.reshape(train_images.shape[0], 28, 28, 1)
    test_images = test_images.reshape(test_images.shape[0], 28, 28, 1)
    train_images = train_images.astype('float32') / 255
    test_images = test_images.astype('float32') / 255

    train_labels = to_categorical(train_labels, 10)
    test_labels = to_categorical(test_labels, 10)
    return (train_images, train_labels), (test_images, test_labels)

模型训练

1. CNN-v1

第一个模型使用使用三个卷积+pooling操作接两个全链接层。

model summary如下：

Model: "sequential_1"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d_1 (Conv2D)            (None, 28, 28, 16)        80        
_________________________________________________________________
max_pooling2d_1 (MaxPooling2 (None, 14, 14, 16)        0         
_________________________________________________________________
conv2d_2 (Conv2D)            (None, 14, 14, 32)        2080      
_________________________________________________________________
max_pooling2d_2 (MaxPooling2 (None, 7, 7, 32)          0         
_________________________________________________________________
conv2d_3 (Conv2D)            (None, 7, 7, 64)          8256      
_________________________________________________________________
max_pooling2d_3 (MaxPooling2 (None, 3, 3, 64)          0         
_________________________________________________________________
dropout_1 (Dropout)          (None, 3, 3, 64)          0         
_________________________________________________________________
flatten_1 (Flatten)          (None, 576)               0         
_________________________________________________________________
dense_1 (Dense)              (None, 500)               288500    
_________________________________________________________________
dropout_2 (Dropout)          (None, 500)               0         
_________________________________________________________________
dense_2 (Dense)              (None, 10)                5010      
=================================================================
Total params: 303,926
Trainable params: 303,926
Non-trainable params: 0
_________________________________________________________________

模型结构如图所示：

最终在测试集上的准确率为：87%

2. CNN-v2

第二个模型比第一个模型更简单，使用了一个卷积+pooling接两个全连接层。

model summary如下：

Model: "sequential_1"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d_1 (Conv2D)            (None, 28, 28, 32)        320       
_________________________________________________________________
max_pooling2d_1 (MaxPooling2 (None, 14, 14, 32)        0         
_________________________________________________________________
flatten_1 (Flatten)          (None, 6272)              0         
_________________________________________________________________
dense_1 (Dense)              (None, 5128)              32167944  
_________________________________________________________________
dense_2 (Dense)              (None, 10)                51290     
=================================================================
Total params: 32,219,554
Trainable params: 32,219,554
Non-trainable params: 0
_________________________________________________________________

模型结构如图所示：

最终在测试集上的准确率为：91.2%

并将预测结果进行显示，图片上名称为红色的为分类错误的图片：

此篇文章为一个系列，未完待续。欢迎大家关注我们的公众号：知识沉淀部落。