# 導(dǎo)入相關(guān)模塊import osimport pandas as pdimport numpy as npimport cv2import matplotlib.pyplot as pltfrom PIL import Imagefrom sklearn.preprocessing import LabelEncoderfrom sklearn.model_selection import train_test_splitimport kerasfrom keras.models import Modelfrom keras.layers import Dense, Activation, Flatten, Dropoutfrom keras.utils import np_utilsfrom keras.applications.resnet50 import ResNet50from?tqdm?import?tqdm

def generate_csv(path):   labels = pd.DataFrame()   # 目錄下每一類別文件夾   items = [f for f in os.listdir(path)]   # 遍歷每一類別文件夾   for i in tqdm(items):   # 生成圖片完整路徑       images = [path + I + '/' + img for img in os.listdir(path+i)]       # 生成兩列:圖像路徑和對應(yīng)標(biāo)簽       labels_data = pd.DataFrame({'images': images, ‘labels’: i})       # 逐條記錄合并       labels = pd.concat((labels, labels_data))   # 打亂順序   labels = labels.sample(frac=1, random_state=42)   return labels
# 生成標(biāo)簽并查看前5行labels = generate_csv('./flowers/')labels.head()

# resize縮放img = cv2.resize(img, (224, 224))# 轉(zhuǎn)換成RGB色彩顯示img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)plt.imshow(img)plt.xticks([])plt.yticks([])

# 定義批量讀取并縮放def read_images(df, resize_dim):   total = 0   images_array = []   # 遍歷標(biāo)簽文件中的圖像路徑   for i in tqdm(df.images):   # 讀取并resize       img = cv2.imread(i)       img_resized = cv2.resize(img, resize_dim)       total += 1       # 存入圖像數(shù)組中       images_array.append(img_resized)   print(total, 'iamges have resized.')   return images_array
# 批量讀取images_array = read_images(labels, (224, 224))

# 轉(zhuǎn)化為圖像數(shù)組X = np.array(images_array)# 標(biāo)簽編碼lbl = LabelEncoder().fit(list(labels['labels'].values))labels['code_labels']=pd.DataFrame(lbl.transform(list(labels['labels'].values)))# 分類標(biāo)簽轉(zhuǎn)換y = np_utils.to_categorical(labels.code_labels.values, 5)

# 劃分為訓(xùn)練和驗證集X_train, X_valid, y_train, y_valid = train_test_split(X,y,test_size=0.2, random_state=42)

# 訓(xùn)練集生成器，中間做一些數(shù)據(jù)增強train_datagen = ImageDataGenerator(   rescale=1./255,   rotation_range=40,   width_shift_range=0.4,   height_shift_range=0.4,   shear_range=0.2,   zoom_range=0.3,   horizontal_flip=True)
# 驗證集生成器，無需做數(shù)據(jù)增強val_datagen = ImageDataGenerator(   rescale=1./255)
# 按批次導(dǎo)入訓(xùn)練數(shù)據(jù)train_generator = train_datagen.flow(   X_train,   y_train,   batch_size=32)
# 按批次導(dǎo)入驗證數(shù)據(jù)val_generator = val_datagen.flow(   X_valid,   y_valid,   batch_size=32)

# 定義模型構(gòu)建函數(shù)def flower_model():   base_model=ResNet50(include_top=False,weights='imagenet', input_shape=(224, 224, 3))   # 凍結(jié)base_model的層，不參與訓(xùn)練       for layers in base_model.layers:       layers.trainable = False   # base_model的輸出并展平   model = Flatten()(base_model.output)   # 添加批歸一化層   model = BatchNormalization()(model)   # 全連接層   model=Dense(2048,activation='relu', kernel_initializer=he_normal(seed=42))(model)   # 添加批歸一化層   model = BatchNormalization()(model)   # 全連接層   model=Dense(1024,activation='relu', kernel_initializer=he_normal(seed=42))(model)   # 添加批歸一化層   model = BatchNormalization()(model)   # 全連接層并指定分類數(shù)和softmax激活函數(shù)   model = Dense(5, activation='softmax')(model)   model = Model(inputs=base_model.input, outputs=model)   # 指定損失函數(shù)、優(yōu)化器、性能度量標(biāo)準(zhǔn)并編譯   model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])   return model

# 調(diào)用模型model = flower_model()# 使用fit_generator方法執(zhí)行訓(xùn)練flower_model.fit_generator(                  generator=train_generator,                  steps_per_epoch=len(train_data)/32,                    epochs=30,                  validation_steps=len(val_data)/32,                  validation_data=val_generator,                  verbose=2                  )