Keras入门-猫狗大战

卷积神经网络被广泛应用于物体检测与分类，本帖从零开始搭建一套基于Keras的猫狗分类卷积网络入门贴。

一、下载猫狗数据集

     下面是我在网络上找到的一个猫狗的数据集，如果不能使用可以在本帖下留言

      链接：https://pan.baidu.com/s/13hw4LK8ihR6-6-8mpjLKDA 密码：dmp4

二、使用数据集

       下载的数据集分为猫狗两类图片，其命名分别以cat、dog命名，我们需要读取文件名信息，将之设置为标签label，便于后续的训练。 下面程序除生成label外，还存在部分代码将图像与标签写入npy文件（方便我更换电脑继续使用训练，免得每次拷贝那么多文件）

def img_to_npy():
    train_idx = 0
    npy_idx   = 0
    path = './/Dataset//kaggle_cats_dogs//train6000/'   #小批量数据集共6000张猫狗的图片
    files = os.listdir(path)
    print("shuffle ing..." )
    random.shuffle(files)                                #打乱数据集图片顺序
    images = []
    labels = []
    print("npy generating ..." )
    for f in files:
        train_idx = train_idx +1
        img_path = path + f
        img = image.load_img(img_path, target_size=image_size)  #image_size = (150, 150)
        img_array = image.img_to_array(img)
        images.append(img_array)

        if 'cat' in f:
            labels.append(0)
        else:
            labels.append(1)

        #***************************************************************************
        if train_idx % npy_size == 0 and train_idx or train_idx == len(files):
            print("train_idx = " + str(train_idx) +";total = " + str(len(files)))
            data = np.array(images)
            labels = np.array(labels)
            labels = np_utils.to_categorical(labels, 2)
            np.save('./Dataset/data_{}.npy'.format(npy_idx), data)
            np.save('./Dataset/labels_{}.npy'.format(npy_idx), labels)
            images = []
            labels = []
            npy_idx = npy_idx + 1
        #***************************************************************************

    print("preprocessing training data done." )
    return data, labels

三、模型网络的搭建

      共四层卷积层、与两个全连接层，二分类

    model = Sequential()

    model.add(Conv2D(32, kernel_size=(5, 5), input_shape=(img_h, img_h, 3), activation='relu', padding='same'))
    model.add(MaxPool2D())
    model.add(Dropout(0.3))

    model.add(Conv2D(64, kernel_size=(5, 5), activation='relu', padding='same'))
    model.add(MaxPool2D())
    model.add(Dropout(0.3))

    model.add(Conv2D(128, kernel_size=(5, 5), activation='relu', padding='same'))
    model.add(MaxPool2D())
    model.add(Dropout(0.5))

    model.add(Conv2D(256, kernel_size=(5, 5), activation='relu', padding='same'))
    model.add(MaxPool2D())
    model.add(Dropout(0.5))

    model.add(Flatten())
    model.add(Dense(512, activation='relu'))
    model.add(Dropout(0.5))

    model.add(Dense(2, activation='softmax'))
    model.summary()

四、训练

我使用自生成的npy文件、（可直接使用归类好的数据集）

    sgd = Adam(lr=0.0003)
    model.compile(loss='binary_crossentropy',optimizer=sgd, metrics=['accuracy'])

    """加载数据"""
    print("img to npy......")
    # img2npy()

    """Tensorboad"""
    tbCallbacks = callbacks.TensorBoard(log_dir='./logs', histogram_freq=1, write_graph=True, write_images=True)

    print("train.......")

    for npy_idx_tmp in range(0, 3):
        print('==============npy_idx_tmp:', npy_idx_tmp)
        images = []
        labels = []
        images = np.load("./Dataset/data_{}.npy".format(npy_idx_tmp))
        labels = np.load("./Dataset/labels_{}.npy".format(npy_idx_tmp))
        # labels = np_utils.to_categorical(labels, 2)
        images /= 255

        x_train, x_test, y_train, y_test = train_test_split(images, labels, test_size=0.2) #随机划分数据集
        print(x_train.shape, y_train.shape)
        model.fit(x_train, y_train, batch_size=nbatch_size, epochs=nepochs, verbose=1, validation_data=(x_test, y_test),
                  callbacks=[tbCallbacks])

        print("evaluate......")
        scroe, accuracy = model.evaluate(x_test, y_test, batch_size=nbatch_size)
        print('scroe:', scroe, 'accuracy:', accuracy)

    yaml_string = model.to_yaml()
    with open('./models/cat_dog.yaml', 'w') as outfile:
        outfile.write(yaml_string)
    model.save_weights('./models/cat_dog.h5')

五、预测

def pred_data():

    with open('./models/cat_dog.yaml') as yamlfile:
        loaded_model_yaml = yamlfile.read()
    model = model_from_yaml(loaded_model_yaml)
    model.load_weights('./models/cat_dog.h5')

    sgd = Adam(lr=0.0003)
    model.compile(loss='categorical_crossentropy',optimizer=sgd, metrics=['accuracy'])

    images = []
    path='./test_images/'
    for f in os.listdir(path):
        img = image.load_img(path + f, target_size=image_size)
        img_array = image.img_to_array(img)

        x = np.expand_dims(img_array, axis=0)
        x = preprocess_input(x)
        result = model.predict_classes(x,verbose=0)

        print(f,result[0])

完整代码：https://download.csdn.net/download/harrison509/11122538