CNN keras 2.1 图片测试

71 阅读 0 评论 47 点赞

我是靠谱客的博主无奈蛋挞，最近开发中收集的这篇文章主要介绍CNN keras 2.1 图片测试，觉得挺不错的，现在分享给大家，希望可以做个参考。

概述

# 设置模型的参数
input_size = row * column # 100*100
batch_size = 32
hidden_neurons = 30
epochs = 25

# 构建一个较为简单的CNN模型

model = Sequential()
model.add(Convolution2D(32, (2, 2), input_shape=(row, column, 1)))
model.add(Activation('relu'))
model.add(Convolution2D(32, (2, 2)))
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.5))
model.add(Flatten())

model.add(Dense(hidden_neurons))
model.add(Activation('relu'))
model.add(Dense(classes))
model.add(Activation('softmax'))

# Define Loss & compile model 在多分类的时候损失函数一般用categorical_crossentropy

使用adadelta 具体选择哪个优化器可以参考官方文档

model.compile(loss='categorical_crossentropy', metrics=['accuracy'], optimizer='adadelta')
# optimizer - "rmsprop"/"sgd"/"adadelta" , loss - "binary_crossentropy" / "categorical_crossentropy"

# fit the model
model.fit(X_train, Y_train, batch_size=batch_size, epochs=epochs, validation_split = 0.1, verbose=1)

测试结果--------------------------------------------------------------------------------------------------------

测试使用的是百度图片中爬取的七种狗类

设置了25个epoch 大概在十个之后会测试成功概率会提高但是在去掉全连接层的概率不太客观

(100, 100, 1)
Car Train data : 567
Total Train Data : 4778
X_train shape
(4778, 100, 100, 1)
Y_train Shape
(4778, 1)

Total Test Data : 520
(4778, 1)
[0]
Train on 4300 samples, validate on 478 samples
Epoch 1/25
4300/4300 [==============================] - 101s 23ms/step - loss: 1.8854 - acc: 0.1737 - val_loss: 2.6176 - val_acc: 0.0000e+00
Epoch 2/25
4300/4300 [==============================] - 100s 23ms/step - loss: 1.6941 - acc: 0.2781 - val_loss: 4.4016 - val_acc: 0.0000e+00
Epoch 3/25
4300/4300 [==============================] - 105s 24ms/step - loss: 1.4562 - acc: 0.4226 - val_loss: 5.0508 - val_acc: 0.0000e+00
Epoch 4/25
4300/4300 [==============================] - 100s 23ms/step - loss: 1.3374 - acc: 0.4847 - val_loss: 4.5513 - val_acc: 0.0000e+00
Epoch 5/25
4300/4300 [==============================] - 101s 23ms/step - loss: 1.2299 - acc: 0.5372 - val_loss: 4.5998 - val_acc: 0.0000e+00
Epoch 6/25
4300/4300 [==============================] - 101s 23ms/step - loss: 1.1201 - acc: 0.5921 - val_loss: 4.9415 - val_acc: 0.0000e+00
Epoch 7/25
4300/4300 [==============================] - 101s 23ms/step - loss: 1.0057 - acc: 0.6374 - val_loss: 4.3453 - val_acc: 0.0000e+00
Epoch 8/25
4300/4300 [==============================] - 100s 23ms/step - loss: 0.8593 - acc: 0.7053 - val_loss: 4.4878 - val_acc: 0.0000e+00
Epoch 9/25
4300/4300 [==============================] - 100s 23ms/step - loss: 0.7286 - acc: 0.7584 - val_loss: 4.5572 - val_acc: 0.0000e+00
Epoch 10/25
4300/4300 [==============================] - 103s 24ms/step - loss: 0.6007 - acc: 0.8105 - val_loss: 4.1045 - val_acc: 0.0084
Epoch 11/25
4300/4300 [==============================] - 102s 24ms/step - loss: 0.4971 - acc: 0.8581 - val_loss: 4.8304 - val_acc: 0.0042
Epoch 12/25
4300/4300 [==============================] - 100s 23ms/step - loss: 0.4063 - acc: 0.8807 - val_loss: 4.6724 - val_acc: 0.0230
Epoch 13/25
4300/4300 [==============================] - 100s 23ms/step - loss: 0.3318 - acc: 0.9100 - val_loss: 4.8855 - val_acc: 0.0356
Epoch 14/25
4300/4300 [==============================] - 103s 24ms/step - loss: 0.2624 - acc: 0.9307 - val_loss: 4.6594 - val_acc: 0.0628
Epoch 15/25
4300/4300 [==============================] - 105s 24ms/step - loss: 0.2086 - acc: 0.9505 - val_loss: 5.5564 - val_acc: 0.0397
Epoch 16/25
4300/4300 [==============================] - 105s 24ms/step - loss: 0.1722 - acc: 0.9563 - val_loss: 6.2152 - val_acc: 0.0418
Epoch 17/25
4300/4300 [==============================] - 105s 24ms/step - loss: 0.1303 - acc: 0.9714 - val_loss: 5.0574 - val_acc: 0.0962
Epoch 18/25
4300/4300 [==============================] - 103s 24ms/step - loss: 0.1002 - acc: 0.9788 - val_loss: 6.6792 - val_acc: 0.0356
Epoch 19/25
4300/4300 [==============================] - 100s 23ms/step - loss: 0.0906 - acc: 0.9800 - val_loss: 6.8854 - val_acc: 0.0335
Epoch 20/25
4300/4300 [==============================] - 100s 23ms/step - loss: 0.0670 - acc: 0.9872 - val_loss: 5.9952 - val_acc: 0.0921
Epoch 21/25
4300/4300 [==============================] - 103s 24ms/step - loss: 0.0567 - acc: 0.9900 - val_loss: 7.9824 - val_acc: 0.0126
Epoch 22/25
4300/4300 [==============================] - 104s 24ms/step - loss: 0.0462 - acc: 0.9935 - val_loss: 8.4218 - val_acc: 0.0146
Epoch 23/25
4300/4300 [==============================] - 104s 24ms/step - loss: 0.0381 - acc: 0.9933 - val_loss: 8.3374 - val_acc: 0.0356
Epoch 24/25
4300/4300 [==============================] - 104s 24ms/step - loss: 0.0320 - acc: 0.9947 - val_loss: 9.1571 - val_acc: 0.0188
Epoch 25/25
4300/4300 [==============================] - 100s 23ms/step - loss: 0.0252 - acc: 0.9956 - val_loss: 6.1695 - val_acc: 0.1423
520/520 [==============================] - 3s 5ms/step
('nTest accuracy:', 0.21153846153846154)