建立CNN网络，对10类图片进行分类判断。
输入：图片大小为32，彩色的，数据为32x32x3，label 10类，进行热独编码
网络：增加了更多的卷积和池化及dropout层
训练：保存模型／重新加载模型（主要是weight）
结果：输出预测值与概率
准确率提高：单纯通过增加epoch数是不能提高准确率的，需要增加更多的卷积层和池化层，训练时间大大曾长。
import numpy as np
import pandas as pd
from keras.utils import np_utils
from keras.datasets import cifar10
from keras.models import Sequential
from keras.layers import Dense
from keras.layers import Dropout, Flatten, Conv2D, MaxPooling2D
import matplotlib.pyplot as plt
import os
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'
np.random.seed(10)
model_save_name = "SavedModel/cifar10CNN.h5"
(x_train_image, y_train_label), (x_test_image, y_test_label) = cifar10.load_data()
print("size train data =", len(x_train_image))
print("size test data =", len(x_test_image))
print(x_train_image.shape)
print(x_train_image[0])
print(y_train_label.shape)
label_dict = {0:"airplane", 1:"automobile", 2:"bird", 3:"cat", 4:"deer",
5:"dog", 6:"frog", 7:"horse", 8:"ship", 9:"truck"}
x_train_normal = x_train_image.astype("float32") / 255.0
x_test_normal = x_test_image.astype("float32") / 255.0
y_train_onehot = np_utils.to_categorical(y_train_label)
y_test_onehot = np_utils.to_categorical(y_test_label)
model = Sequential()
model.add(Conv2D(filters=32,
kernel_size = (3, 3),
padding = 'same',
input_shape = (32, 32, 3),
activation = 'relu'))
model.add(Dropout(0.25))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Conv2D(filters=64,
kernel_size = (3, 3),
padding = 'same',
activation = 'relu'))
model.add(Dropout(0.25))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Flatten())
model.add(Dropout(0.25))
model.add(Dense(units = 1024,
activation = 'relu'))
model.add(Dropout(0.25))
model.add(Dense(units = 10,
activation = 'softmax'))
print(model.summary())
model.compile(loss = "categorical_crossentropy",
optimizer = "adam", metrics = ["accuracy"])
try:
model.load_weights(model_save_name)
except:
print("load model failed!")
history = model.fit(x = x_train_normal,
y = y_train_onehot,
validation_split = 0.2,
epochs = 5,
batch_size = 128,
verbose = 2)
model.save_weights(model_save_name)
def show_train_history(train_history, train, val):
plt.plot(train_history.history[train])
plt.plot(train_history.history[val])
plt.title("Train History")
plt.ylabel(train)
plt.xlabel("Epochs")
plt.legend(["train", "validation"], loc="upper left")
plt.show()
def plot_image_label_prediction(images, labels, prediction, idx = 0, num = 10):
fig = plt.gcf()
fig.set_size_inches(12, 14)
if num > 25:
num = 25
for i in range(0, num):
ax = plt.subplot(5, 5, 1 + i)
ax.imshow(images[idx], cmap="binary")
title = str(i) + "." + label_dict[labels[idx][0]]
if len(prediction) > 0:
title += " => " + label_dict[prediction[idx]]
ax.set_title(title, fontsize = 10)
ax.set_xticks([])
ax.set_yticks([])
idx += 1
plt.show()
def show_prediction_prob(y_label, prediction, x_image, prediction_prob, i):
print("label: ", label_dict[y_label[i][0]], " predict: ", label_dict[prediction[i]])
plt.figure(figsize = (2, 2))
plt.imshow(np.reshape(x_image[i], (32, 32, 3)))
plt.show()
for j in range(10):
print(label_dict[j], " predict probability: %1.9f" % (prediction_prob[i][j]))
show_train_history(history, "acc", "val_acc")
show_train_history(history, "loss", "val_loss")
scores = model.evaluate(x_test_normal, y_test_onehot)
print("accuracy = ", scores[1])
prediction = model.predict_classes(x_test_normal)
print("prediction: ", prediction[:10])
prediction_prob = model.predict(x_test_normal)
plot_image_label_prediction(x_test_image, y_test_label, prediction, idx=0, num=25)
show_prediction_prob(y_test_label, prediction, x_test_image, prediction_prob, 0)
print(pd.crosstab(y_test_label.reshape(-1), prediction, rownames = ["label"], colnames = ["predict"]))
网络结构：
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
conv2d_1 (Conv2D) (None, 32, 32, 32) 896
_________________________________________________________________
dropout_1 (Dropout) (None, 32, 32, 32) 0
_________________________________________________________________
max_pooling2d_1 (MaxPooling2 (None, 16, 16, 32) 0
_________________________________________________________________
conv2d_2 (Conv2D) (None, 16, 16, 64) 18496
_________________________________________________________________
dropout_2 (Dropout) (None, 16, 16, 64) 0
_________________________________________________________________
max_pooling2d_2 (MaxPooling2 (None, 8, 8, 64) 0
_________________________________________________________________
flatten_1 (Flatten) (None, 4096) 0
_________________________________________________________________
dropout_3 (Dropout) (None, 4096) 0
_________________________________________________________________
dense_1 (Dense) (None, 1024) 4195328
_________________________________________________________________
dropout_4 (Dropout) (None, 1024) 0
_________________________________________________________________
dense_2 (Dense) (None, 10) 10250
=================================================================
Total params: 4,224,970
Trainable params: 4,224,970
Non-trainable params: 0
_________________________________________________________________
训练结果：
Train on 40000 samples, validate on 10000 samples
Epoch 1/10
- 103s - loss: 0.3368 - acc: 0.8847 - val_loss: 0.9838 - val_acc: 0.6791
Epoch 2/10
- 105s - loss: 0.2636 - acc: 0.9103 - val_loss: 0.9136 - val_acc: 0.6980
Epoch 3/10
- 102s - loss: 0.6777 - acc: 0.7622 - val_loss: 0.8316 - val_acc: 0.7260
Epoch 4/10
- 108s - loss: 0.5570 - acc: 0.8042 - val_loss: 0.8110 - val_acc: 0.7274
Epoch 5/10
- 108s - loss: 0.4739 - acc: 0.8323 - val_loss: 0.8112 - val_acc: 0.7245
Epoch 6/10
- 97s - loss: 0.4158 - acc: 0.8545 - val_loss: 0.7957 - val_acc: 0.7283
Epoch 7/10
- 103s - loss: 0.3588 - acc: 0.8761 - val_loss: 0.7921 - val_acc: 0.7279
Epoch 8/10
- 103s - loss: 0.3198 - acc: 0.8888 - val_loss: 0.7851 - val_acc: 0.7328
Epoch 9/10
- 98s - loss: 0.2869 - acc: 0.8999 - val_loss: 0.8147 - val_acc: 0.7261
Epoch 10/10
- 100s - loss: 0.2498 - acc: 0.9149 - val_loss: 0.7858 - val_acc: 0.7359
识别结果：
混淆矩阵
predict 0 1 2 3 4 5 6 7 8 9
label
0 758 11 48 12 31 8 11 8 76 37
1 20 796 7 11 9 8 12 3 29 105
2 52 1 649 35 108 53 60 22 10 10
3 22 8 96 470 85 172 91 32 6 18
4 11 1 61 45 763 22 46 37 12 2
5 15 3 76 120 62 639 38 34 4 9
6 1 3 35 44 29 19 861 2 5 1
7 9 1 34 24 80 64 7 770 2 9
8 42 28 18 9 11 3 13 3 848 25
9 34 66 13 12 8 8 8 13 27 811
单个图片的识别结果和概率：
label: cat predict: dog
airplane predict probability: 0.008377961
automobile predict probability: 0.001823489
bird predict probability: 0.016125565
cat predict probability: 0.414337575
deer predict probability: 0.029035321
dog predict probability: 0.430770338
frog predict probability: 0.040058933
horse predict probability: 0.016318403
ship predict probability: 0.036439165
truck predict probability: 0.006713260
问题：为什么要做标准化和归一化，参考：https://blog.csdn.net/resourse_sharing/article/details/51979494
网友评论