-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathsingle_channel_train_cnn_gru.py
128 lines (118 loc) · 5.04 KB
/
single_channel_train_cnn_gru.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
# -*- coding: utf-8 -*-
"""
Created on Fri Nov 2 08:57:02 2018
@author: wen
"""
import numpy as np
#import matplotlib.pyplot as plt
import pandas as pd
import cv2
#from PIL import Image
#from PIL import ImageEnhance
from keras.utils import np_utils
from skimage.transform import resize
from print_loss_accuracy import LossHistory
#from keras.preprocessing.image import load_img, img_to_array
#from sklearn.model_selection import train_test_split
#读入图片名称总文档
data = pd.read_csv('train_cnngru.csv')
data.head(5)
#读取图片信息
path = './train_image_cnngru/'
X = []
count = 1
b = []
for img_name in data.Image_name:
#image = Image.open(path + img_name)
#截取主体图像
#box = (0, 180, 1500, 1080)
#image = image.crop(box)
#灰度图
#image = image.convert('L')
#亮度增强
#enh_bri = ImageEnhance.Brightness(image)
#brightness = 1.5
#image_brightened = enh_bri.enhance(brightness)
#对比度增强
#enh_con = ImageEnhance.Contrast(image_brightened)
#contrast = 1.5
#image_contrasted = enh_con.enhance(contrast)
#image = np.array(image_contrasted)
#image = cv2.fastNlMeansDenoising(image, None, 10, 7, 21)
img = cv2.imread(path + img_name, 0)
equ = cv2.equalizeHist(img) #使图片的像素趋近于高斯分布
image = cv2.GaussianBlur(equ, (5, 5), 0)
a = resize(image, preserve_range = True, output_shape = (224, 224, 1)).astype(int)
a = a / 255 #数据归一化
#ret, binary = cv2.threshold(image, 127, 255, cv2.THRESH_BINARY)
b.append(a)
if count% 5 == 0:
X.append(b)
b = []
count += 1
X = np.array(X)
#独热编码
y = data.Label
dummy_y = np_utils.to_categorical(y) #one hot encoding Classes
y = []
for i in range(0,dummy_y.shape[0]-2,5):
y.append(dummy_y[i:i+5])
y = np.array(y)
#数据集分离
#X_train, X_test, y_train, y_test = train_test_split(X, dummy_y, test_size = 0.2)
X_train = X[250:]
#增加一个维度
#X_train = np.expand_dims(X_train, axis=0)
y_train = y[250:]
X_test = X[:250]
##增加一个维度
#X_test = np.expand_dims(X_test, axis=0)
y_test = y[:250]
#搭建模型
#预处理
from keras.models import Sequential, Model #快速开始序贯模型
from keras.layers import Dense, Flatten, Dropout, Conv2D, MaxPool2D, GRU, Input
#Dence:全连接层（对上一层的神经元进行全部连接，实现特征的非线性组合）
#Flatten返回一个折叠的一维数组
from keras.optimizers import SGD
from keras.layers.wrappers import TimeDistributed
def build_model(categories):
input_shape = Input((224, 224, 1))
cnn = Conv2D(filters=32, kernel_size=3, padding='same', activation='relu')(input_shape)
cnn = MaxPool2D(pool_size= 2, strides = 2)(cnn)
cnn = Conv2D(filters=64, kernel_size=3, padding='same', activation='relu')(cnn)
cnn = MaxPool2D(pool_size= 2, strides = 2)(cnn)
cnn = Conv2D(filters=128, kernel_size=3, padding='same', activation='relu')(cnn)
cnn = MaxPool2D(pool_size= 2, strides = 2)(cnn)
cnn = Conv2D(filters=256, kernel_size=3, padding='same', activation='relu')(cnn)
cnn = MaxPool2D(pool_size= 2, strides = 2)(cnn)
cnn = Conv2D(filters=512, kernel_size=3, padding='same', activation='relu')(cnn)
cnn = MaxPool2D(pool_size= 2, strides = 2)(cnn)
cnn = Flatten()(cnn)
cnn = Dense(512)(cnn)
cnn = Dropout(0.5)(cnn)
x = Model(inputs = input_shape, outputs = cnn)
x.summary()
input_shape = (5, 224, 224, 1) # (seq_len, width, height, channel)
model = Sequential()
model.add(TimeDistributed(x, input_shape=input_shape))
model.add(GRU(512, kernel_initializer='orthogonal', bias_initializer='ones', dropout=0.5, recurrent_dropout=0.5, return_sequences=True))
#model.add(GRU(128, kernel_initializer='orthogonal', bias_initializer='ones', dropout=0.5, recurrent_dropout=0.5, return_sequences=True))
model.add(Dense(categories, activation = 'softmax'))
model.compile(loss='categorical_crossentropy',optimizer = SGD(lr=0.01, momentum=0.9, clipnorm=1., clipvalue=0.5),metrics=['accuracy'])
return model
model = build_model(4)
model.summary()
#编译模型
#SGD：随机梯度下降，随机梯度下降每次的权重更新只利用数据集中的一个样本来完成，momentum表示动量，动量项超参数γ<1一般是小于等于0.9
#加入动量可以使得梯度方向不变的维度上速度变快，梯度方向有所改变的维度上的更新速度变慢，这样就可以加快收敛并减小震荡。
#model.compile(optimizer = SGD(lr=1e-2, momentum = 0.9), loss='categorical_crossentropy', metrics = ['accuracy'])
#model.compile(optimizer = 'adam', loss='categorical_crossentropy', metrics = ['accuracy'])
#记录训练的准确率与损失值
history = LossHistory()
#训练模型
model.fit(X_train, y_train, batch_size = 16, epochs = 100, validation_data = (X_test, y_test), callbacks=[history])
#打印准确率与损失值的变化曲线
history.loss_plot('epoch')
#保存模型
model.save('model_cnn_gru1.h5')