Model Training of Mask Detection II

Import of related libraries

from tensorflow.keras.preprocessing.image import ImageDataGenerator
from tensorflow.keras.applications import MobileNetV2
from tensorflow.keras.layers import AveragePooling2D
from tensorflow.keras.layers import Dropout
from tensorflow.keras.layers import Flatten
from tensorflow.keras.layers import Dense
from tensorflow.keras.layers import Input
from tensorflow.keras.models import Model
from tensorflow.keras.optimizers import Adam
from tensorflow.keras.applications.mobilenet_v2 import preprocess_input
from tensorflow.keras.preprocessing.image import img_to_array
from tensorflow.keras.preprocessing.image import load_img
from tensorflow.keras.utils import to_categorical
from sklearn.preprocessing import LabelBinarizer
from sklearn.model_selection import train_test_split
from sklearn.metrics import classification_report
from imutils import paths
import matplotlib.pyplot as plt
import numpy as np
import argparse
import os

Construct a parameter parser and parse the parameters

• Datasets: input paths for datasets with and without masks
• plot: the path to the output training graph, which will be generated using matplotlib
• Model: the storage path of the generated mask detector model

#path 为你文件夹路径
parser  = argparse.ArgumentParser()
parser .add_argument("-d", "--dataset", required=True, help="path to input dataset",default="path")
parser .add_argument("-p", "--plot", type=str, default="plot.png", help="path to output loss/accuracy plot")
parser .add_argument("-m", "--model", type=str,default="mask_detector.model",help="path to output face mask detector model")
args = vars(parser .parse_args())

Set learning rate, epoch, batch_size

rate = 1e-4
epoch = 20
batch_size = 32

Dataset construction

preprocess_input(), which is similar to a normalization function that comes with keras under tensorflow, which normalizes the incoming image and can speed up the image processing speed. The paths.list_images function can directly obtain the image path under the folder, even if there are subfolders.

imagePaths = list(paths.list_images(args['dataset']))
print(len(imagePaths))
data = []
labels = []
for imagePath in imagePaths:
    print(imagePath)
    label = imagePath.split(os.path.sep)[-2]
    #导入数据，调整size为(224, 224)
    image = load_img(imagePath, target_size=(224, 224))
    image = img_to_array(image)
    # 归一化
    image = preprocess_input(image)

    data.append(data)
    labels.append(labels)

data = np.array(data, dtype= 'float32')
labels = np.array(labels)
# 对图像标签的独热编码
encoder = LabelBinarizer()
labels = encoder.fit_transform(labels)
# 转为二进制
labels = to_categorical(labels)

#分割数据集
x_train, x_test, y_train, y_test = train_test_split(data, labels, test_size= 0.2, stratify= labels, random_state= 42)

data augmentation

ImageDataGenerator() is the image generator in the keras.preprocessing.image module. It can also enhance the data in the batch, expand the size of the data set, and enhance the generalization ability of the model. Such as rotation, deformation, normalization and so on.

• rotation_range(): rotation range
• width_shift_range(): Horizontal translation range
• height_shift_range(): vertical translation range
• zoom_range(): zoom range
• fill_mode: fill mode, constant, nearest, reflect
• horizontal_flip(): horizontal flip
• vertical_flip(): flip vertically

aug = ImageDataGenerator(
	rotation_range=20,
	zoom_range=0.15,
	width_shift_range=0.2,
	height_shift_range=0.2,
	shear_range=0.15,
	horizontal_flip=True,
	fill_mode="nearest"
)

Model fine-tuning

Fine-tuning your settings is a three-step process:

1. Load MobileNet weights pre-trained with ImageNet
2. Construct a new FC header and append it to Baseline to replace the old one
3. Freeze the base layers of the network: During backpropagation, the weights of these base layers will not be updated, while the weights of the head layers will be adjusted

baseModel = MobileNetV2(weights='imagenet', include_top= False, input_tensor= Input(shape=(224, 224,3)))

headModel = baseModel.output
headModel = AveragePooling2D(pool_size = (7, 7)(headModel))
headModel = Flatten(name= 'flatten')(headModel)
headModel = Dense(128, activation= 'relu')(headModel)
headModel = Dropout(0.5)(headModel)
headModel = Dense(2, activation='softmax')(headModel)

model = Model(inputs = baseModel.input, outputs = headModel)
#冻结网络的基本层：在反向传播过程中，这些基本层的权重不会被更新，而头层的权重将被调整
for layer in baseModel.layers:
    layer.trainable = False
#使用Adam优化器、学习率衰减计划和二进制交叉熵编译我们的模型。
optimizer = Adam(lr = rate, decay = rate/epoch )
model.compile(loss="binary_crossentropy", optimizer=optimizer, metrics=["accuracy"])
#数据增强对象（aug）将提供一批经过修改的图像数据
H = model.fit(
	aug.flow(x_train, y_train, batch_size=batch_size),
	steps_per_epoch=len(x_train) // batch_size,
	validation_data=(x_train, y_train),
	validation_steps=len(x_train) // batch_size,
	epochs=x_train)
predIdxs = model.predict(x_test, batch_size=batch_size)

draw loss curve

N = epoch
plt.style.use("ggplot")
plt.figure()
plt.plot(np.arange(0, N), H.history["loss"], label="train_loss")
plt.plot(np.arange(0, N), H.history["val_loss"], label="val_loss")
plt.plot(np.arange(0, N), H.history["accuracy"], label="train_acc")
plt.plot(np.arange(0, N), H.history["val_accuracy"], label="val_acc")
plt.title("Training Loss and Accuracy")
plt.xlabel("Epoch #")
plt.ylabel("Loss/Accuracy")
plt.legend(loc="lower left")
plt.savefig(args["plot"])

Show results: