在线手写识别--ML之Hello World

import tensorflow as tf
# 加载mnist数据集，如果本地没有，会从远程下载到目录~/.keras/datasets/下
mnist = tf.keras.datasets.mnist
(x_train, y_train), (x_test, y_test) = mnist.load_data()

import matplotlib.pyplot as plt
from PIL import Image
import numpy

def plot_samples(samples, labels, rows, cols):
    """
       将样本显示在一张图片上
    """
    fig, axs = plt.subplots(rows, cols, sharex="all", sharey="all")

    axs = axs.flatten()
    for i in range(rows * cols):
        # elem = .reshape(28, 28)
        axs[i].set_title(labels[i], fontsize=10, pad=0)
        # 可以将cmap指定为Greys来展示白底黑字
        axs[i].imshow(samples[i], cmap="gray", interpolation="nearest")
    axs[0].set_xticks([])
    axs[0].set_yticks([])
    plt.tight_layout(h_pad=0.5, pad=0)
    plt.show()
    
# 60000个训练集和10000个测试集
# (60000, 28, 28) (60000,) (10000, 28, 28) (10000,)
print(x_train.shape, y_train.shape, x_test.shape, y_test.shape)
# 显示部分训练集和测试集
plot_samples(x_train, y_train, 10, 10)
plot_samples(x_test, y_test, 10, 10)
# 可以使用PIL提供的方法来展示单张图片
pil_img = Image.fromarray(numpy.uint8(x_train[1]))
pil_img.show()

from keras import layers, models
# 模型定义，3层卷积(Conv2D)池化(MaxPolling2D)+一层全连接(Dense)
model = models.Sequential()
# 第一层需要指定输出的数据尺寸，后面的层通过找推导不需要指定
model.add(layers.Conv2D(32, (3, 3), activation='relu', input_shape=(28, 28, 1)))
model.add(layers.MaxPooling2D((2, 2)))
# 第2层卷积，卷积核大小为3*3，64个
model.add(layers.Conv2D(64, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
# 第3层卷积，卷积核大小为3*3，128个
model.add(layers.Conv2D(128, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Flatten())
model.add(layers.Dense(64, activation='relu'))
model.add(layers.Dense(10, activation='softmax'))

# 编译模型(需要配置学习过程)，可以指定优化函数、损失函数、评估标准
model.compile(
    optimizer=tf.keras.optimizers.legacy.Adam(0.001),
    loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
    metrics=[tf.keras.metrics.SparseCategoricalAccuracy()],
)

import datetime
log_dir = "logs/fit/" + datetime.datetime.now().strftime("%Y%m%d-%H%M%S")
tensorboard_callback = tf.keras.callbacks.TensorBoard(log_dir=log_dir, histogram_freq=1)

# 归一化处理
x_train, x_test = x_train / 255.0, x_test / 255.0
# 对样本进行训练，可以指定迭代次数和批量大小
model.fit(x_train, y_train, epochs=10, callbacks=[tensorboard_callback])
# 将模型保存到指定目录，后续可以直接使用
model.save("./hand_write_model_with_cnn")

model.summary()

# 评估模型效果
model.evaluate(x_test, y_test, verbose=2)

import tensorflow as tf
# 加载训练好的模型
model = tf.keras.models.load_model('./hand_write_model_with_cnn')
print(model.summary())

from PIL import Image,ImageChops,ImageEnhance
import nump

def img_pre_process(file, invert=False, bright=1.5):
    # 对图片进行预处理，因为图片是白底黑字，大小不一，所以需要进行一些预处理
    f = Image.open(file)
    # 转成灰度图片，将原来4维(RGBA)转成一维
    i = f.convert("L")
    # 缩放图片
    i = i.resize((28, 28))
    if invert:
        # 转成黑底白字
        i = ImageChops.invert(i)
    if bright > 0:
        # 按需求加亮，加对比度等
        i = ImageEnhance.Brightness(i).enhance(factor=bright)
    # 转化为numpy数组，shape为28*28
    return numpy.array(i)

# 预处理图片，然后转化为ndarray处理
img9 = img_pre_process("./9.png", bright=0)
samples = [img9 / 255]
img3 = img_pre_process("./3.png", invert=True, bright=0)
samples.append(img3 / 255)
img6 = img_pre_process("./6.png", invert=True)
samples.append(img6 / 255)
img6 = img_pre_process("./8.png", invert=True)
samples.append(img6 / 255)

# 由于predict的输入是批量的，所以这块的shape为n*28*28，n为样本数量
res = model.predict(numpy.array(samples))
# 结果为对各个样本的概率预测结果，以上结果如下
# [[4.5536960e-12 8.6428139e-12 2.2465254e-12 1.1361015e-09 1.4199172e-07
#   1.1165924e-10 1.2581486e-14 3.7364101e-10 3.2622338e-09 9.9999988e-01]
#  [3.4697410e-12 4.4372114e-06 1.4305875e-06 9.9997211e-01 3.2144427e-09
#   3.2603787e-06 1.4439445e-12 1.7748405e-05 5.1269499e-11 9.1026078e-07]
#  [5.4244672e-05 1.5443358e-05 2.2776185e-05 7.1670558e-10 7.0248444e-07
#   2.8083429e-03 9.9709213e-01 4.1633821e-10 4.8526099e-06 1.4761381e-06]
#  [8.4286776e-18 8.2335221e-14 1.3553557e-11 2.5355894e-14 6.0165419e-17
#   2.1572887e-15 1.5686732e-14 9.7833466e-15 1.0000000e+00 1.2415943e-15]]
print(res)

import tensorflow as tf

print("loading model...")
model = tf.keras.models.load_model('../hand_write_model_with_cnn')
print("load model finished.")

from django.http import HttpResponse
from django.shortcuts import render

from PIL import Image, ImageChops
import numpy, io, json, base64
from . import model


def index(request):
    return render(request, 'index.html', {})


def submit(request):
    # canvas图片的数据会转化为base64通过post的data传过来，但这块会有一些前置的信息，需要去除
    base64_str = request.POST['data'].split(",")[1]
    # print(base64_str)

    # base64的图片信息转化为Image
    raw = Image.open(io.BytesIO(base64.decodebytes(bytes(base64_str, "utf-8"))))
    # 由于传过来的image是个PNG，背景是透明的，所以不能直接转灰度，需要将其copy到一张带白底的图上
    img = Image.new('RGB', raw.size, (255, 255, 255))
    img.paste(raw, (0, 0), raw)
    # 然后对图片进行预处理
    img = img.convert("L")
    img = img.resize((28, 28))
    img = ImageChops.invert(img)
    # img.show()

    # 模型预测
    predict = model.predict(numpy.array([numpy.array(img) / 255]))
    # 将处理完的图片转成base64返回
    output_buffer = io.BytesIO()
    img.save(output_buffer, format='JPEG')
    base64_data = str(base64.b64encode(output_buffer.getvalue()), 'utf-8')
    # 同时返回预测结果和处理后的图片
    return HttpResponse(json.dumps(
        (predict.tolist(), base64_data)
    ))

from django.contrib import admin
from django.urls import path

from . import views

urlpatterns = [
    path('admin/', admin.site.urls),
    path('index/', views.index),
    path('submit/', views.submit),
]

import os
MIDDLEWARE = [
    'django.middleware.security.SecurityMiddleware',
    'django.contrib.sessions.middleware.SessionMiddleware',
    'django.middleware.common.CommonMiddleware',
#    'django.middleware.csrf.CsrfViewMiddleware',
    'django.contrib.auth.middleware.AuthenticationMiddleware',
    'django.contrib.messages.middleware.MessageMiddleware',
    'django.middleware.clickjacking.XFrameOptionsMiddleware',
]

ROOT_URLCONF = 'hand_write_site.urls'

TEMPLATES = [
    {
        'BACKEND': 'django.template.backends.django.DjangoTemplates',
        'DIRS': [os.path.join(BASE_DIR, 'templates')],
        'APP_DIRS': True,
        'OPTIONS': {
            'context_processors': [
                'django.template.context_processors.debug',
                'django.template.context_processors.request',
                'django.contrib.auth.context_processors.auth',
                'django.contrib.messages.context_processors.messages',
            ],
        },
    },
]