口罩数据集的划分、训练、测试

这里的方法类似与笑脸识别的方法，也是类似于猫狗识别

可以参考笑脸数据集的划分训练测试

下载数据集

首先我们需要下载口罩数据集
这里我采用的是老师给的数据集 但是里面的数据比较混乱，于是我自己整理了一下，不是口罩的数据集运用到了笑脸数据集里面的图片

划分数据集

import keras
import os, shutil
# The path to the directory where the original
# dataset was uncompressed
original_dataset_dir = 'D:\\mask\\train'
# The directory where we will
# store our smaller dataset
base_dir = 'D:\\mask1'
os.mkdir(base_dir)
# Directories for our training,
# validation and test splits
train_dir = os.path.join(base_dir, 'train')
os.mkdir(train_dir)
validation_dir = os.path.join(base_dir, 'validation')
os.mkdir(validation_dir)
test_dir = os.path.join(base_dir, 'test')
os.mkdir(test_dir)
# Directory with our training masks pictures
train_masks_dir = os.path.join(train_dir, 'masks')
os.mkdir(train_masks_dir)
# Directory with our training unmasks pictures
train_unmasks_dir = os.path.join(train_dir, 'unmasks')
os.mkdir(train_unmasks_dir)
# Directory with our validation masks pictures
validation_masks_dir = os.path.join(validation_dir, 'masks')
os.mkdir(validation_masks_dir)
# Directory with our validation unmasks pictures
validation_unmasks_dir = os.path.join(validation_dir, 'unmasks')
os.mkdir(validation_unmasks_dir)
# Directory with our validation masks pictures
test_masks_dir = os.path.join(test_dir, 'masks')
os.mkdir(test_masks_dir)
# Directory with our validation unmasks pictures
test_unmasks_dir = os.path.join(test_dir, 'unmasks')
os.mkdir(test_unmasks_dir)

建立文件夹之后，自己手动添加进去数据，方法类似笑脸数据集部分

打印查看

print('total training mask images:', len(os.listdir(train_masks_dir)))
print('total training unmask images:', len(os.listdir(train_unmasks_dir)))
print('total validation mask images:', len(os.listdir(validation_masks_dir)))
print('total validation unmask images:', len(os.listdir(validation_unmasks_dir)))
print('total test mask images:', len(os.listdir(test_masks_dir)))
print('total test unmask images:', len(os.listdir(test_unmasks_dir)))

构建小型卷积网络

from keras import layers
from keras import models
model = models.Sequential()
model.add(layers.Conv2D(32, (3, 3), activation='relu',input_shape=(150, 150, 3)))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(64, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(128, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(128, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Flatten())
model.add(layers.Dense(512, activation='relu'))
model.add(layers.Dense(1, activation='sigmoid'))

了解征图的尺寸是如何随着每一层变化的

model.summary()

数据预处理

from keras import optimizersmodel.compile(loss='binary_crossentropy',optimizer=optimizers.RMSprop(lr=1e-4),metrics=['acc'])

from keras.preprocessing.image import ImageDataGenerator# All images will be rescaled by 1./255
train_datagen = ImageDataGenerator(rescale=1./255)
test_datagen = ImageDataGenerator(rescale=1./255)train_generator = train_datagen.flow_from_directory(# This is the target directorytrain_dir,# All images will be resized to 150x150target_size=(150, 150),batch_size=20,# Since we use binary_crossentropy loss, we need binary labelsclass_mode='binary')validation_generator = test_datagen.flow_from_directory(validation_dir,target_size=(150, 150),batch_size=20,class_mode='binary')

输出

for data_batch, labels_batch in train_generator:print('data batch shape:', data_batch.shape)print('labels batch shape:', labels_batch.shape)break

训练

history = model.fit_generator(train_generator,steps_per_epoch=100,epochs=30,validation_data=validation_generator,validation_steps=50)

model.save('D:/mask1/masks_and_unmasks_small_1.h5')

在训练和验证数据上绘制模型的损失和准确性

import matplotlib.pyplot as plt
acc = history.history['acc']
val_acc = history.history['val_acc']
loss = history.history['loss']
val_loss = history.history['val_loss']
epochs = range(len(acc))
plt.plot(epochs, acc, 'bo', label='Training acc')
plt.plot(epochs, val_acc, 'b', label='Validation acc')
plt.title('Training and validation accuracy')
plt.legend()
plt.figure()
plt.plot(epochs, loss, 'bo', label='Training loss')
plt.plot(epochs, val_loss, 'b', label='Validation loss')
plt.title('Training and validation loss')
plt.legend()
plt.show()

数据增强

datagen = ImageDataGenerator(rotation_range=40,width_shift_range=0.2,height_shift_range=0.2,shear_range=0.2,zoom_range=0.2,horizontal_flip=True,fill_mode='nearest')

如果我们使用这种数据增加配置训练一个新的网络，我们的网络将永远不会看到两次相同的输入。然而，它看到的输入仍然是高度相关的，因为它们来自少量的原始图像——我们不能产生新的信息，我们只能混合现有的信息。因此，这可能还不足以完全消除过度拟合。
为了进一步对抗过拟合，我们还将在我们的模型中增加一个Dropout层，就在密集连接分类器之前：

model = models.Sequential()
model.add(layers.Conv2D(32, (3, 3), activation='relu',input_shape=(150, 150, 3)))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(64, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(128, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(128, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Flatten())
model.add(layers.Dropout(0.5))
model.add(layers.Dense(512, activation='relu'))
model.add(layers.Dense(1, activation='sigmoid'))model.compile(loss='binary_crossentropy',optimizer=optimizers.RMSprop(lr=1e-4),metrics=['acc'])

用数据增强 退出来训练我们的网络

train_datagen = ImageDataGenerator(rescale=1./255,rotation_range=40,width_shift_range=0.2,height_shift_range=0.2,shear_range=0.2,zoom_range=0.2,horizontal_flip=True,)
# Note that the validation data should not be augmented!
test_datagen = ImageDataGenerator(rescale=1./255)
train_generator = train_datagen.flow_from_directory(# This is the target directorytrain_dir,# All images will be resized to 150x150target_size=(150, 150),batch_size=32,# Since we use binary_crossentropy loss, we need binary labelsclass_mode='binary')
validation_generator = test_datagen.flow_from_directory(validation_dir,target_size=(150, 150),batch_size=32,class_mode='binary')
history = model.fit_generator(train_generator,steps_per_epoch=100,epochs=50,validation_data=validation_generator,validation_steps=50)

由于训练时间太长了，我把这里改成了50，就是训练50次

但是训练时间依然很长

保存模型

model.save('D:/mask1/masks_and_unmasks_small_2.h5')

在训练和验证数据上绘制模型的损失和准确性

acc = history.history['acc']
val_acc = history.history['val_acc']
loss = history.history['loss']
val_loss = history.history['val_loss']
epochs = range(len(acc))
plt.plot(epochs, acc, 'bo', label='Training acc')
plt.plot(epochs, val_acc, 'b', label='Validation acc')
plt.title('Training and validation accuracy')
plt.legend()
plt.figure()
plt.plot(epochs, loss, 'bo', label='Training loss')
plt.plot(epochs, val_loss, 'b', label='Validation loss')
plt.title('Training and validation loss')
plt.legend()
plt.show()

但是由于训练时间太长了，我就不进一步训练了

实现摄像头口罩识别判断

基于口罩数据集训练出的模型进行识别

#检测视频或者摄像头中的人脸
import cv2
from keras.preprocessing import image
from keras.models import load_model
import numpy as np
import dlib
from PIL import Image
model = load_model('D:/mask1/masks_and_unmasks_small_2.h5')
detector = dlib.get_frontal_face_detector()
video=cv2.VideoCapture(0)
font = cv2.FONT_HERSHEY_SIMPLEX
def rec(img):gray=cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)dets=detector(gray,1)if dets is not None:for face in dets:left=face.left()top=face.top()right=face.right()bottom=face.bottom()cv2.rectangle(img,(left,top),(right,bottom),(0,255,0),2)img1=cv2.resize(img[top:bottom,left:right],dsize=(150,150))img1=cv2.cvtColor(img1,cv2.COLOR_BGR2RGB)img1 = np.array(img1)/255.img_tensor = img1.reshape(-1,150,150,3)prediction =model.predict(img_tensor)    if prediction[0][0]>0.5:result='unmask'else:result='mask'cv2.putText(img, result, (left,top), font, 2, (0, 255, 0), 2, cv2.LINE_AA)cv2.imshow('Video', img)
while video.isOpened():res, img_rd = video.read()if not res:breakrec(img_rd)if cv2.waitKey(5) & 0xFF == ord('q'):break
video.release()
cv2.destroyAllWindows()

测试如下：

通过开源的数据模型进行识别

下载

开源人脸口罩检测模型和数据

这个模型是基于tensorflow和keras的，要先配置好

运行

然后直接运行这个文件就可以了

我是在Spyder里面运行的，也是在anaconda环境下面，对于这种文件夹我感觉更好操作一点