-
Notifications
You must be signed in to change notification settings - Fork 1
/
FF-BLStrain.py
142 lines (119 loc) · 5.17 KB
/
FF-BLStrain.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
129
130
131
132
133
134
135
136
137
138
139
140
141
142
import numpy as np
import os
import cv2
from BroadLearningSystem import BLStrain, BLS_AddEnhanceNodes, BLS_AddFeatureEnhanceNodes, bls_train_input, \
bls_train_inputenhance
import keras
def transformLabel(label_raw):######################################################此时类别是两个类
#results = [0, 0]
if label_raw == 1: ##
results = [0, 1]
#elif label_raw == '0':
else:
results = [1, 0]
return results
def rotateImg(img_raw):
rows, cols = img_raw.shape[:2]
result_img = []
for i in range(10):
M = cv2.getRotationMatrix2D((cols / 2, rows / 2), i * 5, 1)
dest_img = cv2.warpAffine(img_raw, M, (rows, cols))
result_img.append(dest_img)
M2 = cv2.getRotationMatrix2D((cols / 2, rows / 2), -i * 5, 1)
dest_img2 = cv2.warpAffine(img_raw, M2, (rows, cols))
result_img.append(dest_img2)
return result_img
#######卷积层处理#####
from keras.models import Sequential,Model,Sequential
from keras.layers import Input, Dense, Dropout, Flatten, Conv2D, MaxPooling2D
from keras.applications.vgg16 import VGG16
from keras.applications.resnet50 import ResNet50
############通过txt文件进行数据读取，txt存储有图片的路径和类别
def getDatanew(filePath, resize_format=(100, 100), colorType=cv2.IMREAD_COLOR, resize_interpolation=cv2.INTER_NEAREST,
needGenImg=False):####IMREAD_COLOR IMREAD_GRAYSCALE
tmpData = []
tmpLabel = []
##
model = ResNet50(include_top=True, weights='G:/zy/01/resnet50_weights_tf_dim_ordering_tf_kernels.h5')
dense_result = Model(inputs=model.input, outputs=model.get_layer("avg_pool").output) ####分类层的前一层
with open(filePath) as txtData:
lines = txtData.readlines()
for line in lines:
file, label = line.split(' ')
#print(label)
label=int(label)####将string转换为int
tmpLabel.append(transformLabel(label))】
fileName = file
img = cv2.imread(fileName, colorType)
img_formated = cv2.resize(img, resize_format, interpolation=resize_interpolation)
img_formated = np.expand_dims(img_formated, axis=0)
img_flat=dense_result.predict(img_formated)
img_flat = img_flat.ravel()
tmpData.append(img_flat)
if needGenImg:
genImgs = rotateImg(img_formated)####通过旋转图片进行数据集增广
for genImg in genImgs:
img_flat = img_formated.ravel()
tmpData.append(img_flat)
tmpLabel.append(transformLabel(label))
return np.double(tmpData), np.double(tmpLabel)####返回样本和标签
import time
# 获取数据
filePath_train = './crack/train.txt'
filePath_test = './crack/test.txt'
resize_format = (224, 224)####(128, 128)
t1=time.time()
traindata, trainlabel = getDatanew(filePath_train, resize_format, colorType=cv2.IMREAD_COLOR,needGenImg=False)##不需要旋转图片 #####cv2.IMREAD_COLOR False
t2=time.time()
traindatatime=t2-t1
print("traindatatime is:",traindatatime)
t3=time.time()
testdata, testlabel = getDatanew(filePath_test, resize_format, colorType=cv2.IMREAD_COLOR,needGenImg=False) ###不需要旋转图片 #####cv2.IMREAD_GRAYSCALE
t4=time.time()
testdatatime=t4-t3
print("testdatatime is:",testdatatime)
# 设置参数
N1 = 9 # # of nodes belong to each window 5 {'N1': 2.0, 'N2': 64.0, 'N3': 18278.0
N2 = 32 # # of windows -------Feature mapping layer 7 94 6 48 25514
N3 = 2083 # # of enhancement nodes -----Enhance layer 200 2083
L = 6 # # of incremental steps
M1 = 500 # # of adding enhance nodes
s = 0.8 # shrink coefficient
C = 2 ** -30 # Regularization coefficient
#print(testlabel.shape)###
#print(trainlabel.shape)###
#print(testlabel)
# 跑BLS
print('-------------------BLS_BASE---------------------------')#'N1': 13.0, 'N2': 30.0, 'N3': 26518.0
BLStrain(traindata, trainlabel, testdata, testlabel, s, C, N1, N2, N3)
#################################BayesianOptimization################################
# from bayes_opt import BayesianOptimization
# from hyperopt import fmin, tpe, hp, STATUS_OK, Trials
# import hyperopt.pyll.stochastic
#
# def BLSopt(argsDict):
#
# N1 = int(argsDict["N1"])
# N2 = int(argsDict['N2'])
# N3 = int(argsDict["N3"])
#
# test_acc=BLStrain(traindata, trainlabel,testdata, testlabel, s, C, N1, N2, N3)
#
# return -test_acc
#
# spaceBL = {
# 'N1': hp.quniform('N1', 1,10,1),
# 'N2': hp.quniform('N2', 1,60,2),
# 'N3': hp.quniform('N3', 1000,2500,2)
# }
#
# #print(hyperopt.pyll.stochastic.sample(space4knn))
#
# trials = Trials()
# best = fmin(BLSopt, space=spaceBL, algo=tpe.suggest, max_evals=10, trials=trials)
# print('best:',best)
########################################################################################
#Incremental learning
# M1 = 2000 # # of adding new patterns
# print('-------------------BLS_INPUT--------------------------')
# bls_train_input(traindata[0:20000,:],trainlabel[0:20000,:],traindata,trainlabel,testdata,testlabel,s,C,N1,N2,N3,L,M1)