kerasによる最終層の出力の設定

###問題点

kerasによる画像認識を拝見し，Inception v3のfine tuning(4クラス分類)をやりたいと思っています．
しかし，最終層の設定において出力を4つに設定し，モデルのサマリを見ても

dense_1 (Dense)                  (None, 4)             8196        global_average_pooling2d_1[0][0]

となっているにも関わらず，以下のエラーが出ます

ValueError: Error when checking target: expected dense_1 to have shape (None, 1) but got array with shape (32, 4)

何が原因かが未だ分からないので助言をお願いしたいと思います．
以下ソースコードです．
###該当のソースコード

Python
1import os
2from keras.applications.inception_v3 import InceptionV3
3from keras.applications.inception_v3 import preprocess_input
4from keras.models import Sequential, Model
5from keras.layers import Dense, Dropout, Activation, Flatten
6from keras.layers import Convolution2D, MaxPooling2D, ZeroPadding2D, GlobalAveragePooling2D, AveragePooling2D
7from keras.regularizers import l2
8from keras.utils import np_utils
9from keras.utils import plot_model
10from keras.preprocessing.image import ImageDataGenerator
11from keras.callbacks import ModelCheckpoint
12import matplotlib.pyplot as plt
13import keras.backend as K
14
15epoch = 2
16
17result_dir = 'D:/result'
18if not os.path.exists(result):
19    os.mkdir(result_dir)
20
21
22def plot_history(history):
23    # 精度の履歴をプロット
24    plt.plot(history.history['acc'], "o-", label="accuracy")
25    plt.plot(history.history['val_acc'], "o-", label="val_acc")
26    plt.title('model accuracy')
27    plt.xlabel('epoch')
28    plt.ylabel('accuracy')
29    plt.legend(loc="lower right")
30    plt.show()
31
32    # 損失の履歴をプロット
33    plt.plot(history.history['loss'], "o-", label="loss", )
34    plt.plot(history.history['val_loss'], "o-", label="val_loss")
35    plt.title('model loss')
36    plt.xlabel('epoch')
37    plt.ylabel('loss')
38    plt.legend(loc='lower right')
39    plt.show()
40
41
42def save_history(history, result_file):
43    loss = history.history['loss']
44    acc = history.history['acc']
45    val_loss = history.history['val_loss']
46    val_acc = history.history['val_acc']
47    nb_epoch = len(acc)
48
49    with open(result_file, "w") as fp:
50        fp.write("epoch\tloss\tacc\tval_loss\tval_acc\n")
51        for i in range(nb_epoch):
52            fp.write("%d\t%f\t%f\t%f\t%f\n" % (i, loss[i], acc[i], val_loss[i], val_acc[i]))
53
54
55if __name__ == '__main__':
56    classes = ['m', 'n', 'o', 'k']
57
58    batch_size = 32
59    nb_classes = len(classes)
60    img_rows, img_cols = 299, 299
61
62    samples_per_epoch = 600
63    nb_val_samples = 99
64
65    # CNNを構築
66    # Inception v3モデルの読み込み，最終層は読み込まない
67    base_model = InceptionV3(weights='imagenet', include_top=False)
68    # 最終層の設定
69    x = base_model.output
70    x = GlobalAveragePooling2D()(x)
71    predictions = Dense(nb_classes, kernel_initializer="glorot_uniform", activation="softmax",
72                        kernel_regularizer=l2(.0005))(x)
73
74    # モデルのサマリを表示
75    model = Model(inputs=base_model.input, outputs=predictions)
76    plot_model(model, show_shapes=True, to_file=os.path.join(result_dir, 'model.png'))
77    model.summary()
78
79    # base_modelはweightsを更新しない
80    for layer in base_model.layers:
81        layer.trainable = False
82
83    model.compile(loss='categorical_crossentropy',
84                  optimizer='adam',
85                  metrics=['accuracy'])
86
87    # ディレクトリの画像を使ったジェネレータ
88    train_datagen = ImageDataGenerator(
89        rescale=1.0 // 255,
90        shear_range=0.2,
91        zoom_range=0.2,
92        horizontal_flip=True)
93
94    test_datagen = ImageDataGenerator(rescale=1.0 // 255)
95
96    train_generator = train_datagen.flow_from_directory(
97        directory='D:/s_dir/train_images',
98        target_size=(img_rows, img_cols),
99        color_mode='rgb',
100        classes=classes,
101        class_mode='categorical',
102        batch_size=batch_size,
103        shuffle=True)
104
105    # 確認
106    print(train_generator.class_indices)
107
108    test_generator = test_datagen.flow_from_directory(
109        directory='D:/s_dir/test_images',
110        target_size=(img_rows, img_cols),
111        color_mode='rgb',
112        classes=classes,
113        class_mode='categorical',
114        batch_size=batch_size,
115        shuffle=True)
116
117    checkpointer = ModelCheckpoint(filepath='D:/result/model.{epoch:02d}-{val_loss:.2f}.hdf5', verbose=1,
118                                   save_best_only=True)
119
120    history = model.fit_generator(
121        train_generator,
122        steps_per_epoch=samples_per_epoch,
123        epochs=epoch,
124        validation_data=test_generator,
125        validation_steps=nb_val_samples,
126        callbacks=[checkpointer])
127
128    save_history(history, os.path.join(result, 'history.txt'))
129
130    # modelに学習させた時の変化の様子をplot
131    plot_history(history)
132
133    loss, acc = model.evaluate_generator(test_generator, steps=400)
134    print('Test loss:', loss)
135    print('Test acc:', acc)
136
137    K.clear_session()
138