keras 簡單的分類實例

這段程序是使用keras進行深度學習分類問題。比較簡單，將每個類別單獨保存在一個文件夾中，然後包含所有類別的文件夾在放在同一個文件夾下面。比如下面所示：

train/

        cat/

        dag/

下面的代碼寫的是9分類的問題，

代碼：

from keras.applications.inception_v3 import InceptionV3
from keras.preprocessing import image
from keras.models import Model
from keras.layers import Dense, GlobalAveragePooling2D
from keras import backend as K
import keras
from keras.preprocessing.image import  ImageDataGenerator
from keras.callbacks import ModelCheckpoint, LearningRateScheduler
from keras.callbacks import ReduceLROnPlateau
from keras.optimizers import SGD
from keras.models import Sequential, load_model
from keras.layers import Activation, Dropout, Flatten, Dense, GlobalMaxPooling2D, BatchNormalization, Input, Conv2D
from keras.applications.inception_v3 import InceptionV3
from keras import metrics
from keras.optimizers import Adam 



import math
import os
import numpy as np 
import pandas as pd
import glob
from tqdm import tqdm
import cv2

num_classes = 9
image_height = 139
image_weight = 139
batch_size = 16
epochs = 50

def train_gen():
    train_datagen = ImageDataGenerator(
            rescale=1./255,
            shear_range=0.2,
            zoom_range=0.2,
            horizontal_flip=True)
    train_generator = train_datagen.flow_from_directory(
        'data/train',
        target_size=(image_height, image_weight),
        batch_size=32,
        class_mode='categorical')

    return train_generator


def test_gen():
    test_datagen = ImageDataGenerator(rescale=1./255)
    test_generator = test_datagen.flow_from_directory(
            './data/test',
            target_size=(image_height, image_weight),
            batch_size=32,
            class_mode='categorical')
    return test_generator

# 計算訓練集的數量
def train_num_samples():
    path = './data/train' # 文件夾下有9個類別，所以這裏統計9個類別，所有的圖片數量
    count = 0
    for i in os.listdir(path):
        pt = os.path.join(path, i)
        ls = glob.glob(os.path.join(pt, '*.jpg'))
        count += len(ls)
    return count

train_generator = train_gen()


def create_model(input_shape, n_out):
    input_tensor = Input(shape=input_shape)
    base_model = InceptionV3(include_top=False,
                   weights='imagenet',
                   input_shape=input_shape)
    bn = BatchNormalization()(input_tensor)
    x = base_model(bn)
    x = Conv2D(32, kernel_size=(1,1), activation='relu')(x)
    x = Flatten()(x)
    x = Dropout(0.5)(x)
    x = Dense(1024, activation='relu')(x)
    x = Dropout(0.5)(x)
    output = Dense(n_out, activation='softmax')(x)
    
    model = Model(input_tensor, output)
    
    return model

# # create the base pre-trained model
# base_model = InceptionV3(weights='imagenet', include_top=False)



# warm up model
model = create_model(
    input_shape=(image_height,image_weight,3), 
    n_out=num_classes)
# we need to recompile the model for these modifications to take effect
# we use SGD with a low learning rate

model.compile(optimizer=SGD(lr=0.0001, momentum=0.9), loss='categorical_crossentropy', metrics = ['accuracy'])


model_checkpoint = ModelCheckpoint('./model.hdf5', monitor='loss', save_best_only=True, mode='auto', verbose = 1)
reduce_lr = ReduceLROnPlateau(monitor='loss', patience=15, mode='auto') 

model_callbacks = [model_checkpoint, reduce_lr]
model.fit_generator(
        train_generator,
        steps_per_epoch=math.ceil( train_num_samples() / batch_size),
        epochs=epochs,
        verbose = 1,
        callbacks = model_callbacks)



# 預測部分
# 加載模型
model.load_weights('./model.hdf5')

def predict(MODEL):
    paths = glob.glob(os.path.join('./data/test', '*.jpg'))
    count = 0
    ID = []
    pro1 = []
    pro2 = []
    pro3 = []
    pro4 = []
    pro5 = []
    pro6 = []
    pro7 = []
    pro8 = []
    pro9 = []

    for i in tqdm(paths):
        # if count > 3:
            # break
        img = cv2.imread(i)
        img = cv2.resize(img, (image_height, image_weight))
        pre = MODEL.predict(np.expand_dims(img/255., 0))[0]
        # print(type(pre))
        # print(pre.shape)
        pre = pre.reshape((1,9))
        ID.append(i.split(os.sep)[-1])
        pro1.append(pre[0][0])
        pro2.append(pre[0][1])
        pro3.append(pre[0][2])
        pro4.append(pre[0][3])
        pro5.append(pre[0][4])
        pro6.append(pre[0][5])
        pro7.append(pre[0][6])
        pro8.append(pre[0][7])
        pro9.append(pre[0][8])

        # count += 1
    d = {'ID':ID, '001':pro1,'002':pro2,'003':pro3,'004':pro4,'005':pro5,'006':pro6,'007':pro7,'008':pro8,'009':pro9}
    # print(d)
    df = pd.DataFrame(d)
    df.to_csv('./result.csv', index = False, encoding='utf-8')
    

predict(model)