机器学习——是否带口罩识别
（一）选题背景：2019年以来，新冠状病毒的出现，给人们带来了沉重的生命财产的损失。我国对肺炎疫情期间也采取了严肃的措施，在公共场合要求人们必须全程佩戴口罩。佩戴口罩可以更好的起到预防病症出现爆发传播的情况，口罩可以有效防止细菌和病毒的入侵。并且还可以起到防止飞沫的扩散。并且新型冠状病毒感染和呼吸道飞沫传播是传播的主要途径，戴口罩可有效防止病情传播。因此我们要通过机器学习方式识别图中的人是否佩戴了口罩。
（二）机器学习设计案例设计方案：网络中下载相应的数据集并分组创建出新的数据集，在python的环境中，查看各组数据集数量，对数据进行预处理，利用keras，构建网络，训练模型，观察训练次数的精度变化， 导入图片测试模型
数据集来源：csnd，网址：数据来源入口
（三）机器学习的实现步骤：
1.下载数据集：
2.导入需要用到的库：
1 import os 2 import numpy as np 3 import os.path 4 import matplotlib.pyplot as plt 5 from PIL import Image 6 from keras import layers 7 from keras import models 8 from keras import optimizers 9 from keras.preprocessing.image import ImageDataGenerator 10 from keras.models import load_model 11 from keras.utils import image_utils
3.查看各组数据集数量：
1 train_path="E:/Jupyter_files/mask_and_nomask_small/train/"
2 print('total training mask images:', len(os.listdir(train_path+"mask")))
3 print('total training nomask images:', len(os.listdir(train_path+"nomask")))
4 valid_path="E:/Jupyter_files/mask_and_nomask_small/validation/"
5 print('total validation mask images:', len(os.listdir(valid_path+"mask")))
6 print('total validation nomask images:', len(os.listdir(valid_path+"nomask")))
7 test_path="E:/Jupyter_files/mask_and_nomask_small/test/"
8 print('total test images:', len(os.listdir(test_path)))
4.搭建网络并查看特征图的维度如何随着每层变化
1 model = models.Sequential() 2 #第一个卷积层作为输入层，32个3*3卷积核，输入形状input_shape = (150,150,3) 3 # 输出图片尺寸：150-3+1=148*148，参数数量：32*3*3*3+32=896 4 model.add(layers.Conv2D(32,(3,3),activation = 'relu',input_shape = (150,150,3))) 5 model.add(layers.MaxPooling2D((2,2)))# 输出图片尺寸：148/2=74*74 6 #输出图片尺寸：74-3+1=72*72，参数数量：64*3*3*32+64=18496 7 model.add(layers.Conv2D(64,(3,3),activation = 'relu')) 8 model.add(layers.MaxPooling2D((2,2)))# 输出图片尺寸：72/2=36*36 9 # 输出图片尺寸：36-3+1=34*34，参数数量：128*3*3*64+128=73856 10 model.add(layers.Conv2D(128,(3,3),activation = 'relu')) 11 model.add(layers.MaxPooling2D((2,2)))# 输出图片尺寸：34/2=17*17 12 model.add(layers.Conv2D(128,(3,3),activation = 'relu')) 13 model.add(layers.MaxPooling2D((2,2))) 14 model.add(layers.Flatten()) 15 model.add(layers.Dense(512,activation = 'relu')) 16 model.add(layers.Dense(1,activation = 'sigmoid'))#sigmoid分类，输出是两类别 17 #看一下特征图的维度如何随着每层变化 18 model.summary()
5.搭建模型建立数据和验证数据
1 from keras import optimizers
2 #图像在输入神经网络之前进行数据处理，建立训练和验证数据
3 from keras.preprocessing.image import ImageDataGenerator
4 model.compile(loss='binary_crossentropy',
5 optimizer=optimizers.RMSprop(lr=1e-4),
6 metrics=['acc'])
7 #归一化
8 train_datagen = ImageDataGenerator(rescale = 1./255)
9 test_datagen = ImageDataGenerator(rescale = 1./255)
10 train_dir = 'E:/Jupyter_files/mask_and_nomask_small/train' #指向训练集图片目录路径
11 train_generator = train_datagen.flow_from_directory(
12 train_dir,
13 target_size = (150,150),# 输入训练图像尺寸
14 batch_size = 20,
15 class_mode = 'binary') #
16 validation_dir = 'E:/Jupyter_files/mask_and_nomask_small/validation' #指向验证集图片目录路径
17 validation_generator = test_datagen.flow_from_directory(
18 validation_dir,
19 target_size = (150,150),
20 batch_size = 20,
21 class_mode = 'binary')
22 for data_batch,labels_batch in train_generator:
23 print('data batch shape:',data_batch.shape)
24 print('data batch shape:',labels_batch.shape)
25 break #生成器不会停止，会循环生成这些批量，所以我们就循环生成一次批量
6.训练模型并保存
1 history = model.fit(
2 train_generator,
3 steps_per_epoch = 100,
4 epochs = 50,
5 validation_data = validation_generator,
6 validation_steps = 50)
7 #将训练过程产生的数据保存为h5文件8 model.save('E:/Jupyter_files/new_mask_and_nomask_10epoch1.h5')
6.绘制Epochs与Accuracy变化曲线图
%matplotlib inline
epochs = range(10) #50 is the number of epochs
train_acc = history.history['acc']
valid_acc = history.history['val_acc']
plt.plot(epochs, train_acc, 'bo',label = 'Training Accuracy')
plt.plot(epochs, valid_acc, 'r', label = 'Validation Accuracy')
plt.xlabel('Epochs')
plt.ylabel('Accuracy')
plt.legend()
plt.show()
7.读取一样本观察特征图
7.1显示样本
img_path = "E:/Jupyter_files/mask_and_nomask_small/test/1220.png" img = image_utils.load_img(img_path, target_size=(150,150)) img_tensor = image_utils.img_to_array(img) img_tensor = np.expand_dims(img_tensor, axis=0) img_tensor /= 255. print(img_tensor.shape) #显示样本 import matplotlib.pyplot as plt plt.imshow(img_tensor[0]) plt.show()
7.2获得改变样本的特征图并显示第一层第五个特征图
layer_outputs = [layer.output for layer in model.layers[:8]] activation_model = models.Model(inputs=model.input, outputs=layer_outputs) #获得改样本的特征图 activations = activation_model.predict(img_tensor) first_layer_activation = activations[0] plt.matshow(first_layer_activation[0,:,:,5], cmap="viridis")
7.3显示前2层的全部特征图
1 layer_names = []
2 for layer in model.layers[:2]:
3 layer_names.append(layer.name)
4 # 每行显示16个特征图
5 images_pre_row = 16 #每行显示的特征图数
6 # 循环8次显示8层的全部特征图
7 for layer_name, layer_activation in zip(layer_names, activations):
8 n_features = layer_activation.shape[-1] #保存当前层的特征图个数
9 size = layer_activation.shape[1] #保存当前层特征图的宽高
10 n_col = n_features // images_pre_row #计算当前层显示多少行
11 #生成显示图像的矩阵
12 display_grid = np.zeros((size*n_col, images_pre_row*size))
13 #遍历将每个特张图的数据写入到显示图像的矩阵中
14 for col in range(n_col):
15 for row in range(images_pre_row):
16 #保存该张特征图的矩阵(size,size,1)
17 channel_image = layer_activation[0,:,:,col*images_pre_row+row]
18 #为使图像显示更鲜明，作一些特征处理
19 channel_image -= channel_image.mean()
20 channel_image /= channel_image.std()
21 channel_image *= 64
22 channel_image += 128
23 #把该特征图矩阵中不在0-255的元素值修改至0-255
24 channel_image = np.clip(channel_image, 0, 255).astype("uint8")
25 #该特征图矩阵填充至显示图像的矩阵中
26 display_grid[col*size:(col+1)*size, row*size:(row+1)*size] = channel_image
27 scale = 1./size
28 #设置该层显示图像的宽高
29 plt.figure(figsize=(scale*display_grid.shape[1],scale*display_grid.shape[0]))
30 plt.title(layer_name)
31 plt.grid(False)
32 #显示图像
33 plt.imshow(display_grid, aspect="auto", cmap="viridis")
8.读取测试像样本，改变其尺寸
1 def convertjpg(jpgfile,outdir,width=150,height=150): #将图片缩小到（150，150）的大小
2 img=Image.open(jpgfile)
3 try:
4 new_img=img.resize((width,height),Image.BILINEAR)
5 new_img.save(os.path.join(outdir,os.path.basename(new_file)))
6 except Exception as e:
7 print(e)
8 jpgfile="E:/Jupyter_files/mask_and_nomask_small/test/0023.jpg"
9 new_file="E:/Jupyter_files/mask_and_nomask_small/19.jpg"
10 #图像大小改变到（150,150），文件名保存
11 convertjpg(jpgfile,r"E:/Jupyter_files/mask_and_nomask_small")
12 img_scale = plt.imread('E:/Jupyter_files/mask_and_nomask_small/19.jpg')
13 plt.imshow(img_scale)
9.导入测试样本图片进行预测
1 model = load_model('E:/Jupyter_files/new_mask_and_nomask_10epoch1.h5')
2 img_scale = plt.imread('E:/Jupyter_files/mask_onmask_small/19.jpg')
3 img_scale = img_scale.reshape(1,150,150,3).astype('float32')
4 img_scale = img_scale/255 #归一化到0-1之间
5 out= model.predict(img_scale) #取图片信息
6 img_scale = plt.imread('E:/Jupyter_files/mask_onmask_small/19.jpg')
7 plt.imshow(img_scale) #显示图片
8 if out>0.5:
9 print('该图片是带口罩的概率为：',out)
10 else:
11 print('该图片是不带口罩的概率为：',1-out)
全部代码附上：
1 #需要导入的库
2 import os
3 import numpy as np
4 import os.path
5 import matplotlib.pyplot as plt
6 from PIL import Image
7 from keras import layers
8 from keras import models
9 from keras import optimizers
10 from keras.preprocessing.image import ImageDataGenerator
11 from keras.models import load_model
12 from keras.utils import image_utils
13 #检查分组中包含多少张图像
14 train_path="E:/Jupyter_files/mask_and_nomask_small/train/"
15 print('total training mask images:', len(os.listdir(train_path+"mask")))
16 print('total training nomask images:', len(os.listdir(train_path+"nomask")))
17 valid_path="E:/Jupyter_files/mask_and_nomask_small/validation/"
18 print('total validation mask images:', len(os.listdir(valid_path+"mask")))
19 print('total validation nomask images:', len(os.listdir(valid_path+"nomask")))
20 test_path="E:/Jupyter_files/mask_and_nomask_small/test/"
21 print('total test images:', len(os.listdir(test_path)))
22 model = models.Sequential()
23 #第一个卷积层作为输入层，32个3*3卷积核，输入形状input_shape = (150,150,3)
24 # 输出图片尺寸：150-3+1=148*148，参数数量：32*3*3*3+32=896
25 model.add(layers.Conv2D(32,(3,3),activation = 'relu',input_shape = (150,150,3)))
26 model.add(layers.MaxPooling2D((2,2)))# 输出图片尺寸：148/2=74*74
27 #输出图片尺寸：74-3+1=72*72，参数数量：64*3*3*32+64=18496
28 model.add(layers.Conv2D(64,(3,3),activation = 'relu'))
29 model.add(layers.MaxPooling2D((2,2)))# 输出图片尺寸：72/2=36*36
30 # 输出图片尺寸：36-3+1=34*34，参数数量：128*3*3*64+128=73856
31 model.add(layers.Conv2D(128,(3,3),activation = 'relu'))
32 model.add(layers.MaxPooling2D((2,2)))# 输出图片尺寸：34/2=17*17
33 model.add(layers.Conv2D(128,(3,3),activation = 'relu'))
34 model.add(layers.MaxPooling2D((2,2)))
35 model.add(layers.Flatten())
36 model.add(layers.Dense(512,activation = 'relu'))
37 model.add(layers.Dense(1,activation = 'sigmoid'))#sigmoid分类，输出是两类别
38 #看一下特征图的维度如何随着每层变化
39 model.summary()
40 #图像在输入神经网络之前进行数据处理，建立训练和验证数据
41 from keras.preprocessing.image import ImageDataGenerator
42 model.compile(loss='binary_crossentropy',
43 optimizer=optimizers.RMSprop(lr=1e-4),
44 metrics=['acc'])
45 #归一化
46 train_datagen = ImageDataGenerator(rescale = 1./255)
47 test_datagen = ImageDataGenerator(rescale = 1./255)
48 train_dir = 'E:/Jupyter_files/mask_and_nomask_small/train' #指向训练集图片目录路径
49 train_generator = train_datagen.flow_from_directory(
50 train_dir,
51 target_size = (150,150),# 输入训练图像尺寸
52 batch_size = 20,
53 class_mode = 'binary') #
54 validation_dir = 'E:/Jupyter_files/mask_and_nomask_small/validation' #指向验证集图片目录路径
55 validation_generator = test_datagen.flow_from_directory(
56 validation_dir,
57 target_size = (150,150),
58 batch_size = 20,
59 class_mode = 'binary')
60 for data_batch,labels_batch in train_generator:
61 print('data batch shape:',data_batch.shape)
62 print('data batch shape:',labels_batch.shape)
63 break #生成器不会停止，会循环生成这些批量，所以我们就循环生成一次批量
64 #训练模型50轮次
65 history = model.fit(
66 train_generator,
67 steps_per_epoch = 100,
68 epochs = 50,
69 validation_data = validation_generator,
70 validation_steps = 50)
71 #将训练过程产生的数据保存为h5文件
72 from keras.models import load_model
73 model.save('E:/Jupyter_files/new_mask_and_nomask_10epoch1.h5')
74 #打印每个epoch的accuracy
75 %matplotlib inline
76 epochs = range(10) #50 is the number of epochs
77 train_acc = history.history['acc']
78 valid_acc = history.history['val_acc']
79 plt.plot(epochs, train_acc, 'bo',label = 'Training Accuracy')
80 plt.plot(epochs, valid_acc, 'r', label = 'Validation Accuracy')
81 plt.xlabel('Epochs')
82 plt.ylabel('Accuracy')
83 plt.legend()
84 plt.show()
85 #从测试集中读取一条样本
86 img_path = "E:/Jupyter_files/mask_and_nomask_small/test/1220.png"
87 img = image_utils.load_img(img_path, target_size=(150,150))
88 img_tensor = image_utils.img_to_array(img)
89 img_tensor = np.expand_dims(img_tensor, axis=0)
90 img_tensor /= 255.
91 print(img_tensor.shape)
92 #显示样本
93 plt.imshow(img_tensor[0])
94 plt.show()
95 layer_outputs = [layer.output for layer in model.layers[:8]]
96 activation_model = models.Model(inputs=model.input, outputs=layer_outputs)
97 #获得改样本的特征图
98 activations = activation_model.predict(img_tensor)
99 first_layer_activation = activations[0]
100 plt.matshow(first_layer_activation[0,:,:,5], cmap="viridis")
101 layer_names = []
102 for layer in model.layers[:2]:
103 layer_names.append(layer.name)
104 # 每行显示16个特征图
105 images_pre_row = 16 #每行显示的特征图数
106 for layer_name, layer_activation in zip(layer_names, activations):
107 n_features = layer_activation.shape[-1] #保存当前层的特征图个数
108 size = layer_activation.shape[1] #保存当前层特征图的宽高
109 n_col = n_features // images_pre_row #计算当前层显示多少行
110 #生成显示图像的矩阵
111 display_grid = np.zeros((size*n_col, images_pre_row*size))
112 #遍历将每个特张图的数据写入到显示图像的矩阵中
113 for col in range(n_col):
114 for row in range(images_pre_row):
115 #保存该张特征图的矩阵(size,size,1)
116 channel_image = layer_activation[0,:,:,col*images_pre_row+row]
117 #为使图像显示更鲜明，作一些特征处理
118 channel_image -= channel_image.mean()
119 channel_image /= channel_image.std()
120 channel_image *= 64
121 channel_image += 128
122 #把该特征图矩阵中不在0-255的元素值修改至0-255
123 channel_image = np.clip(channel_image, 0, 255).astype("uint8")
124 #该特征图矩阵填充至显示图像的矩阵中
125 display_grid[col*size:(col+1)*size, row*size:(row+1)*size] = channel_image
126 scale = 1./size
127 #设置该层显示图像的宽高
128 plt.figure(figsize=(scale*display_grid.shape[1],scale*display_grid.shape[0]))
129 plt.title(layer_name)
130 plt.grid(False)
131 #显示图像
132 plt.imshow(display_grid, aspect="auto", cmap="viridis")
133 def convertjpg(jpgfile,outdir,width=150,height=150): #将图片缩小到（150，150）的大小
134 img=Image.open(jpgfile)
135 try:
136 new_img=img.resize((width,height),Image.BILINEAR)
137 new_img.save(os.path.join(outdir,os.path.basename(new_file)))
138 except Exception as e:
139 print(e)
140 jpgfile="E:/Jupyter_files/mask_and_nomask_small/image_mask/0023.jpg"
141 new_file="E:/Jupyter_files/mask_and_nomask_small/19.jpg"
142 #图像大小改变到（150,150），文件名保存
143 convertjpg(jpgfile,r"E:/Jupyter_files/mask_and_nomask_small")
144 img_scale = plt.imread('E:/Jupyter_files/mask_and_nomask_small/19.jpg')
145 plt.imshow(img_scale)
146 model = load_model('E:/Jupyter_files/new_mask_and_nomask_10epoch1.h5')
147 img_scale = plt.imread('E:/Jupyter_files/mask_and_nomask_small/19.jpg')
148 img_scale = img_scale.reshape(1,150,150,3).astype('float32')
149 img_scale = img_scale/255 #归一化到0-1之间
150 out= model.predict(img_scale) #取图片信息
151 img_scale = plt.imread('E:/Jupyter_files/mask_and_nomask_small/19.jpg')
152 plt.imshow(img_scale) #显示图片
153 if out>0.5:
154 print('该图片是带口罩的概率为：',out)
155 else:
156 print('该图片是带口罩的概率为：',1-out)
标签：150,口罩,plt,机器,img,mask,model,识别,nomask From： https://www.cnblogs.com/wszw/p/16989435.html