代码功能: 读取高光谱mat文件，按照领域窗口进行分割,制作数据集和标签集
Ps:是分割······
import numpy as np
import scipy.io as scio
import matplotlib.pyplot as plt
import math
from sklearn.cross_validation import train_test_split
import operator
#用于将一维索引转换成二维索引，制作label图
def chage_to_matrix_index(num,height,weight):
temp_index = num
if int(temp_index) % int(height) != 0:
temp_col = math.floor(temp_index / height)
temp_row = int(math.fmod(temp_index, height) - 1)
else:
temp_col = math.floor(temp_index / height)-1
temp_row = int(height - 1)
return temp_row,temp_col
#裁剪后的数据块的标签，去对应的label小块图，得出种类
def matrix_make_label(dataset,Y_cell,X_cell):
classCount={}
for i in range(Y_cell):
for j in range(X_cell):
tt = dataset[i][j]
if tt not in classCount.keys(): classCount[ tt ] = 0
classCount[ tt ] += 1
sortedClassCount = sorted(classCount.items(), key=operator.itemgetter(1), reverse=True)
#1-只有0 2-有0最多,且两种即以上 3-0不是最多的
if (sortedClassCount[0][0] ==0)and(len(sortedClassCount)==1):
return 999 #代表不应该参与分类
elif (sortedClassCount[0][0] ==0)and(len(sortedClassCount)!=1):
return sortedClassCount[1][0]
else:
return sortedClassCount[0][0]
def get_files():
data_mat = scio.loadmat(“Salinas.mat”)
img = np.array(data_mat[‘img’]) # read data from dict
GroundT = np.array(data_mat[‘GroundT’]).T # read data from dict(54129, 2)
[height,weight,channels] = img.shape
print(height,weight,channels)#512 217 204
img_GroundT = np.zeros([height,weight])
#print(GroundT.shape[0])
#制作label构成的图
for t in range(GroundT.shape[0]):
i,j = chage_to_matrix_index(GroundT[t][0],height,weight)#取出索引，转换为下标
img_GroundT[i][j] = GroundT[t][1]
#print(GroundT[0][0],GroundT[0][1]) 33 15
#print(chage_to_matrix_index(GroundT[0][0],row,col)) (32, 0)
#print(img_GroundT[32][0]) 15.0
#矩阵元素[row,col]
#=====================================label图制作完成=====================
X_cell = 29;
Y_cell = 29;
label_list = []
train_list = []
#判断划分后的块label值有没有用,标记999
img_useful = np.zeros([math.floor(height/Y_cell),math.floor(weight/X_cell)])
#先裁剪出label图,先裁行再裁列
for i in range(math.floor(height/Y_cell)): #(i+1)*29 -1 不会超height
start = i*Y_cell #i为row-1 0*29 1*29
last = (i+1)*Y_cell # 1*29 2*29
dataCol_h = img_GroundT[start:last,:]
#Path_ = strcat(path,'_Label_','row',num2str(row));
for j in range(math.floor(weight/X_cell)): #(i+1)*29 -1 不会超weight
start = j*X_cell #j为col-1 0*29 1*29
last = (j+1)*X_cell # 1*29 2*29
dataCol_x = dataCol_h[:,start:last]
#=============================
#对矩阵操作法①转为one-hot想家多数表决，法②转为字典，封装为函数
#print(dataCol_x.shape)
temp = matrix_make_label(dataCol_x,Y_cell,X_cell)
#最后列表append的应该是个label值
#===============
label_list.append(temp)
if temp == 999:
img_useful[i][j] = 999
#print('img_useful',img_useful)
#print(np.array(label_list).shape)
#print(label_list[87])
#ok~但此时label还有999
for i in range(math.floor(height/Y_cell)): #(i+1)*29 -1 不会超height
start = i*Y_cell #i为row-1 0*29 1*29
stop = (i+1)*Y_cell-1 # 1*29 -1 2*29 -1
dataCol_h = img[start:stop,:,:]
#Path_ = strcat(path,'_Label_','row',num2str(row));
for j in range(math.floor(weight/X_cell)): #(i+1)*29 -1 不会超weight
start = j*X_cell #j为col-1 0*29 1*29
stop = (j+1)*X_cell-1 # 1*29 -1 2*29 -1
#==================================================
#检测label值是否为0，为0就不加这个train块,总共7*17(119)块里有些全是0的····
if img_useful[i][j] == 999:
continue
#==================================================
dataCol_x = dataCol_h[:,start:stop,:]
train_list.append(dataCol_x)
#print(np.array(train_list).shape)
#print(train_list[0].shape)
#print(train_list[0][0][0])
#print(train_list[0])
#ok~
#=========================================
#剔除完train块里的0的，在剔除label的999的
#逻辑：label和train都是遍历切行再切列，顺序一致
#区别只是label在序列上多余了999，所以直接删除值为999的即可
re_label_list = []
for value in label_list:
if value != 999:
re_label_list.append(value)
#print("len----label_list",len(label_list))
#print("len----re_label_list",len(re_label_list))#=========剔除完也是92个
#print(label_list)
#print(re_label_list)
#====================制作完train_list和label_list====================
return train_list,re_label_list
if name == ‘main‘:
train_list,re_label_list = get_files()
print(np.array(train_list).shape)
print(np.array(re_label_list).shape)