树莓派基于tensorflow的数字识别

树莓派基于tensorflow的数字识别

目前博主只试过python3.7.3+tensorflow1.13.1版本，其它tensorflow版本的还没试

一、安装tensorflow环境

• 检查python环境

python3 --version
pip3 --version

• 更新软件源

sudo apt update
sudo apt upgrade

• 执行安装相应环境

sudo apt-get install python3-pip python3-dev

• 若使用pip3 install tensorflow==1.13.1安装可能需要等待漫长的时间，可以先在网上下载 tensorflow-1.13.1-cp37-none-linux_armv7l.whl，然后复制到树莓派系统上，再执行以下代码可以安装成功

sudo pip3 install tensorflow-1.13.1-cp37-none-linux_armv7l.whl

• 运行以下程序检验环境是否安装成功

import tensorflow as tf
hello = tf.constant(“Hello, World!”)
sess = tf.Session()
print(sess.run(hello))

• 解决numpy和h5py的依赖

sudo apt install libatlas-base-dev
sudo apt install libhdf5-dev
sudo apt install python-h5py

• 安装numpy和h5py

sudo pip3 install h5py
sudo pip3 install numpy

• 配置opencv2环境

sudo apt-get install build-essential cmake git pkg-config
sudo apt install build-essential cmake git pkg-config libgtk-3-dev libcanberra-gtk*
sudo apt install libavcodec-dev libavformat-dev libswscale-dev libv4l-dev libxvidcore-dev libx264-dev
sudo apt install libjpeg-dev libpng-dev libtiff-dev gfortran openexr libatlas-base-dev opencl-headers
sudo apt install python3-dev python3-numpy libtbb2 libtbb-dev libdc1394-22-dev
sudo apt-get install libjpeg8-dev
sudo apt-get install libtiff5-dev
sudo apt-get install libjasper-dev
sudo apt-get install libpng12-dev
sudo apt-get install libavcodec-dev libavformat-dev libswscale-dev libv4l-dev
sudo apt-get install libgtk2.0-dev
sudo apt-get install libatlas-base-dev gfortran

• 安装opencv2

// 下载OpenCV
sudo apt-get install libopencv-dev
sudo apt-get install python-opencv

 

二、程序分析

• 项目文件结构，生成模型放置在model文件夹，测试数据集在testimage，训练数据集在trainimage，mnist.py为训练模型程序，main.py为部署模型程序

• mnist.py部分程序分析
• 导入库（使用tensorflow框架）

# -*- coding: UTF-8 -*-
import tensorflow as tf
import tensorflow.keras as keras
from tensorflow.keras import Sequential
from tensorflow.keras.layers import Flatten, Dense, Conv2D, MaxPooling2D, Dropout, MaxPool2D
from tensorflow.keras.datasets import mnist
from tensorflow.keras import backend as K
import json

 

• 设置参数（batch_size为每次训练所选取的样本数，epochs为训练次数）

# 设置参数
batch_size = 128
num_classes = 10
epochs = 10
 
# 输入数据维度
img_rows, img_cols = 28, 28

• •  加载数据集

# 加载数据集，调用库中已有mnist数据集from tensorflow.keras.datasets import mnist
(x_train, y_train), (x_test, y_test) = mnist.load_data()

if K.image_data_format() == 'channels_first':
    x_train = x_train.reshape(x_train.shape[0], 1, img_rows, img_cols)
    x_test = x_test.reshape(x_test.shape[0], 1, img_rows, img_cols)
    input_shape = (1, img_rows, img_cols)
else:
    x_train = x_train.reshape(x_train.shape[0], img_rows, img_cols, 1)
    x_test = x_test.reshape(x_test.shape[0], img_rows, img_cols, 1)
    input_shape = (img_rows, img_cols, 1)
    
x_train = x_train.astype('float32') / 255
x_test = x_test.astype('float32') / 255

• 构造神经网络模型（损失函数使用交叉熵损失函数，优化器使用adam，衡量模型指标为准确率）

# 构建网络
model = Sequential()
# 第一个卷积层，32个卷积核，大小５x5，卷积模式SAME,激活函数relu,输入张量的大小
model.add(Conv2D(filters=32, kernel_size=(5, 5), padding='Same', activation='relu',
                 input_shape=(28, 28, 1)))
model.add(Conv2D(filters=32, kernel_size=(5, 5), padding='Same', activation='relu'))
# 池化层,池化核大小２x2
model.add(MaxPool2D(pool_size=(2, 2)))
# 随机丢弃四分之一的网络连接，防止过拟合
model.add(Dropout(0.25))
model.add(Conv2D(filters=64, kernel_size=(3, 3), padding='Same', activation='relu'))
model.add(Conv2D(filters=64, kernel_size=(3, 3), padding='Same', activation='relu'))
model.add(MaxPool2D(pool_size=(2, 2), strides=(2, 2)))
model.add(Dropout(0.25))
# 全连接层,展开操作，
model.add(Flatten())
# 添加隐藏层神经元的数量和激活函数
model.add(Dense(256, activation='relu'))
model.add(Dropout(0.25))
# 输出层
model.add(Dense(10, activation='softmax'))
model.summary()
model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])

• • 模型训练

model.fit(x_train, y_train, batch_size=batch_size,epochs=epochs,verbose=1,validation_data=(x_test, y_test))
score = model.evaluate(x_test, y_test, verbose=0)
print('Test loss:', score[0])
print('Test accuracy:', score[1])

• 模型及其参数保存（保存格式为h5）

with open('model.json', 'w') as outfile:
    json.dump(model.to_json(), outfile)
 
model_file = 'model.h5'
model.save(model_file)

 

• 部署模型程序分析
• 导入库（numpy、tensorflow、matplotlib、cv2、picamera）

# -*- coding: UTF-8 -*-
import numpy as np
import tensorflow as tf
import tensorflow.keras as keras
from tensorflow.keras.preprocessing.image import img_to_array, load_img
import matplotlib.pyplot as plt
import matplotlib.image as mpimg
from PIL import Image
import cv2
import os
from picamera import PiCamera
from time import sleep

 

• 加载模型model.h5

#load model
model_file = './model/model.h5'
model_file = model_file
print(type(model_file))
global model
model = keras.models.load_model(model_file)

 

• 调用picamera库来连接树莓派的摄像头，并通过摄像头拍摄一张分辨率为480*480的图片，将其保存至“/home/pi/Desktop/camera/tf_keras_mnist/image_28.jpg”

# 调用打开摄像头库
camera = PiCamera()
# 设置照片分辨率为480*480
camera.resolution = (480, 480)
camera.start_preview()
sleep(2)
camera.capture('/home/pi/Desktop/camera/tf_keras_mnist/image_28.jpg')
camera.stop_preview()
print("Collect Image Finish!")
img_file = './image_28.jpg'

 

• 读取“./image_28.jpg”位置的图片，并将其分辨率更改为28*28，使得满足模型参数输入要求

img_array = cv2.imread(img_file, cv2.IMREAD_GRAYSCALE)
resized_image = cv2.resize(img_array, (28, 28))
 
img = Image.fromarray(np.uint8(resized_image))
img.save('output.jpg')
img = mpimg.imread(img_file)

 

• 把28*28分辨率的图片通过img_to_array把其矩阵中的参数由整数值变为浮点数的数组，再把所得数组放进已训练好的模型中，最后将会输出一个预测值

img = img_to_array(load_img(img_file, target_size=(28, 28), color_mode="grayscale")) / 255.
img = np.expand_dims(img, axis=0)
code = model.predict_classes(img)[0]
print("Predict Result: ", code)

 

 全代码区

# -*- coding: UTF-8 -*-
import numpy as np
import tensorflow as tf
import tensorflow.keras as keras
from tensorflow.keras.preprocessing.image import img_to_array, load_img
import matplotlib.pyplot as plt
import matplotlib.image as mpimg
from PIL import Image
import cv2
import os
from picamera import PiCamera
from time import sleep

#load model
model_file = './model/model.h5'
model_file = model_file
print(type(model_file))
global model
model = keras.models.load_model(model_file)


def preditc():
    print("Get ready to capture images and place the camera")
    count = 5
    while count >= 1:
        print("Count Down: ", count, "s")
        count = count - 1
        sleep(1)

    # 调用打开摄像头库
    camera = PiCamera()
    # 设置照片分辨率为480*480
    camera.resolution = (480, 480)
    camera.start_preview()
    sleep(2)
    camera.capture('/home/pi/Desktop/camera/tf_keras_mnist/image_28.jpg')
    camera.stop_preview()
    print("Collect Image Finish!")
    img_file = './image_28.jpg'

    img_array = cv2.imread(img_file, cv2.IMREAD_GRAYSCALE)
    resized_image = cv2.resize(img_array, (28, 28))

    img = Image.fromarray(np.uint8(resized_image))
    img.save('output.jpg')
    img = mpimg.imread(img_file)

    img = img_to_array(load_img(img_file, target_size=(28, 28), color_mode="grayscale")) / 255.
    img = np.expand_dims(img, axis=0)
    code = model.predict_classes(img)[0]
    print("Predict Result: ", code)

    plt.imshow(np.real(img).squeeze())
    plt.show()


# 主函数
if __name__ == '__main__':
    preditc()