练习题目--裂缝识别

#练习题目–裂缝识别
1、数据集准备
数据集为混凝土照片，共分为两类，存在裂缝的图片和不存在裂缝的图片。整个数据集分为训练集、验证集、测试集三部分，其比例可以不同，不同的比例可能造成不同的训练效果，这里练习过程中采用训练集、验证集、测试集：3:1:1。；注意：所有图片的尺寸需要保持一致，这里是128×128的图片，可以根据实际情况自己选择，图像越大运算速度越慢。下图是存在裂缝的照片

2、数据读取及处理准备工作
（1）算法库导入

from keras.layers import Conv2D,MaxPooling2D,Dropout,Dense,Flatten,Activation
import numpy as np
from keras.layers.normalization import BatchNormalization
import cv2
from keras.preprocessing.image import ImageDataGenerator
import matplotlib.pyplot as plt
from keras.utils import plot_model
from keras import optimizers
import os
from    keras.utils.vis_utils    import plot_model
import keras
from keras import regularizers
from keras.callbacks import Callback
from keras.models import Sequential
from keras.callbacks import ModelCheckpoint,EarlyStopping#检查点

（2）读取图片数据
这里是二分类，类别设置为2、每次对128张图片进行批量训练，

num_classes = 2 #类别
batch_size =128 #批量大小
epochs =50#训练次数
#读取图片数据
file1=os.listdir('F:/BankCardOCR/dataset/digit_dataSet/train(128^2)/1')
file2=os.listdir('F:/BankCardOCR/dataset/digit_dataSet/train(128^2)/0')
file3=os.listdir('F:/BankCardOCR/dataset/digit_dataSet/validation128^2/1')
file4=os.listdir('F:/BankCardOCR/dataset/digit_dataSet/validation128^2/0')
m=len(file1)+len(file2)#训练集图片数量
m2=len(file3)+len(file4)#验证集图片数量

（3）处理工作

def changeDim(img):#扩展维度变4维
    img=np.expand_dims(img,axis=2)#
    return img
def Generator(path, batch_size):
    data = []
    label = []
    while True:
        file=os.listdir(path)
        i=0
        img1=[]
        img2=[]
        imgname1=os.listdir(path + '/' + file[0])
        imgname2 = os.listdir(path + '/' + file[1])
        for name1 in imgname1:
            img1.append(path + '/' + file[0]+'/'+name1)
        for name2 in imgname2:
            img2.append(path + '/' + file[1] + '/'+name2)
        imgname=img1+img2######横向连接
        # print(imgname)
        print((len(imgname)))
        np.random.shuffle(imgname)
        #读取图像，根据文件名制作标签
        for finame in imgname:
            im = cv2.imread(finame,0)
            label_a=finame.split('/')[-2]
            im=changeDim(im)
            data.append(im)
            label.append(label_a)
            if (len(label) == batch_size ):
                data = np.array(data)
                data=data.astype('float32')
                data/=255.0#归一化
                label = keras.utils.to_categorical(label, 2)#标签热码
                yield data, label
                data = []
                label = []
            i+=1

3、网络结构设计

model = createModel(num_classes)
def createModel(num_classes):
    model = Sequential()  # 顺序模型
    model.add(
        Conv2D(16, (3, 3), strides=(1, 1), padding="same", input_shape=(128, 128, 1), data_format='channels_last',
               kernel_initializer='uniform', activation="relu"))
    # model.add(BatchNormalization())
    model.add(MaxPooling2D(pool_size=(2, 2), strides=(2, 2)))
    # model.add(Activation('relu'))

    model.add(Conv2D(32, (3, 3), strides=(1, 1), activation="relu"))
    # model.add(BatchNormalization())
    # model.add(Activation('relu'))
    model.add(MaxPooling2D(pool_size=(2, 2), strides=(2, 2)))

    model.add(Conv2D(64, (3, 3), activation="relu"))
    model.add(Conv2D(128, (3, 3), activation="relu"))
    model.add(MaxPooling2D(pool_size=(2, 2), strides=(2, 2), name="pool5"))
    model.add(Conv2D(256, (1, 1)))
    model.add(Conv2D(256, (2, 2)))
    model.add(Conv2D(256, (2, 2), activation="relu"))
    model.add(MaxPooling2D(pool_size=(2, 2), strides=(2, 2), name="pool1"))
    model.add(Dropout(0.25))
    model.add(Flatten())
    model.add(Dense(256))
    model.add(BatchNormalization())
    model.add(Activation('relu'))
    model.add(BatchNormalization())
    model.add(Dropout(0.5))
    model.add(Dense(num_classes))
    model.add(BatchNormalization())
    model.add(Activation("softmax"))
    model.summary()
    return model

4、运行

model.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])#此处二分类选用'binary_crossentropy'损失函数
# 设置检查点
checkpointer = ModelCheckpoint(filepath='./checkpoint/weights(s128).hdf5', monitor='val_loss',verbose=1, save_best_only=True)
#训练集图片路径
trainpath='F:/BankCardOCR/dataset/digit_dataSet/train(128^2)'
#测试集图片路径
testpath='F:/BankCardOCR/dataset/digit_dataSet/validation128^2'
history=model.fit_generator(Generator(trainpath, batch_size),
                    steps_per_epoch=int(m) // batch_size,
                    epochs=epochs,
                    verbose=1,
                    validation_data=Generator(testpath, batch_size),
                    validation_steps=int(m2) // batch_size,
                    shuffle=True,
                    callbacks=[checkpointer])
# 保存训练模型
model.save('./checkpoint/model(s128).h5')
通过测试集进行验证
score = model.evaluate_generator(Generator(testpath,batch_size),steps=int(m) // batch_size)
print('validation loss:', score[0])
print('validation accuracy:', score[1])
print(history.history.keys())

5、模型训练结果
训练结果

plt.plot(history.history['acc'])
plt.plot(history.history['val_acc'])
plt.title('model1 accuracy')
plt.ylabel('accuracy')
plt.xlabel('epoch')
plt.legend(['trian','test'],loc='upper left')
plt.show()

plt.plot(history.history['loss'])
plt.plot(history.history['val_loss'])
plt.title('model1 loss')
plt.ylabel('loss')
plt.xlabel('epoch')
plt.legend(['trian','test'],loc='upper left')
plt.show()