该代码支持多图像裁剪
先将tif格式的图片转为png
再对多个png图片进行批量裁剪
批量裁剪：
import os
# import gdal_makeData
import numpy as np
from osgeo import gdal
os.environ["OPENCV_IO_MAX_IMAGE_PIXELS"] = pow(2,40).__str__()
import cv2 as cv
from PIL import Image
from PIL import ImageFile
ImageFile.LOAD_TRUNCATED_IMAGES = True
Image.MAX_IMAGE_PIXELS = None
'''
single_cutImg函数的imread可以将输入的png格式的图片转为8位彩色后，再进行裁剪
'''
def read_gdal(path):
"""
读取一个tiff图像
:param path: 要读取的图像路径(包括后缀名)
:type path: string
:return im_data: 返回图像矩阵(h, w, c)
:rtype im_data: numpy
:return im_proj: 返回投影信息
:rtype im_proj: ?
:return im_geotrans: 返回坐标信息
:rtype im_geotrans: ?
"""
image = gdal.Open(path) # 打开该图像
if image is None:
print(path + "文件无法打开")
return
img_w = image.RasterXSize # 栅格矩阵的列数
img_h = image.RasterYSize # 栅格矩阵的行数
im_bands = image.RasterCount # 波段数
im_proj = image.GetProjection() # 获取投影信息
im_geotrans = image.GetGeoTransform() # 仿射矩阵
im_data = image.ReadAsArray(0, 0, img_w, img_h)
# 二值图一般是二维，需要添加一个维度
if len(im_data.shape) == 2:
im_data = im_data[np.newaxis, :, :]
im_data = im_data.transpose((1, 2, 0))
print("栅格矩阵的列数: ", img_w)
print("栅格矩阵的行数: ", img_h)
print("栅格矩阵的波段数: ", im_bands)
print("栅格矩阵的投影信息: ", im_proj)
print("栅格矩阵的仿射矩阵信息: ", im_geotrans)
print("ia_data形状：", im_data.shape)
return im_data, im_proj, im_geotrans
def write_gdal(im_data, path, im_proj=None, im_geotrans=None):
"""
重新写一个tiff图像
:param im_data: 图像矩阵(h, w, c)
:type im_data: numpy
:param im_proj: 要设置的投影信息(默认None)
:type im_proj: ?
:param im_geotrans: 要设置的坐标信息(默认None)
:type im_geotrans: ?
:param path: 生成的图像路径(包括后缀名)
:type path: string
:return: None
:rtype: None
"""
im_data = im_data.transpose((2, 0, 1))
if 'int8' in im_data.dtype.name:
datatype = gdal.GDT_Byte
elif 'int16' in im_data.dtype.name:
datatype = gdal.GDT_UInt16
elif 'float32' in im_data.dtype.name:
datatype = gdal.GDT_Float32
else:
datatype = gdal.GDT_Float64
if len(im_data.shape) == 3:
im_bands, im_height, im_width = im_data.shape
elif len(im_data.shape) == 2:
im_data = np.array([im_data])
else:
im_bands, (im_height, im_width) = 1, im_data.shape
# 创建文件
driver = gdal.GetDriverByName("GTiff")
dataset = driver.Create(path, im_width, im_height, im_bands, datatype)
if dataset is not None:
if im_geotrans is None or im_proj is None:
pass
else:
dataset.SetGeoTransform(im_geotrans) # 写入仿射变换参数
dataset.SetProjection(im_proj) # 写入投影
for i in range(im_bands):
dataset.GetRasterBand(i + 1).WriteArray(im_data[i])
del dataset
def single_set_proj_trans(ori_path, target_path):
"""
为 target_path 影像设置 ori_path 影像的投影、坐标信息
:param ori_path: 获取 ori_path 影像路径
:type ori_path: string
:param target_path: 获取 target_path 影像路径
:type target_path: string
:return: None
:rtype: None
"""
# 原图像导入
_, im_proj, im_geotrans = read_gdal(ori_path)
# 目标二值图导入
im_data, _, _ = read_gdal(target_path)
# 地理信息写入
write_gdal(im_data, target_path, im_proj, im_geotrans)
def tif_bands_math(tif_data):
# 读取image.tif 4bands 转换为uint16 3bands
h, w, bands = tif_data.shape
print("获取函数传递的tif数据为：", tif_data.shape)
band1 = tif_data[:, :, 0]
band2 = tif_data[:, :, 1]
band3 = tif_data[:, :, 2]
# band4 = tif_data[:, :, 3]
# band5 = tif_data[:, :, 4]
# band6 = tif_data[:, :, 5]
# band7 = tif_data[:, :, 6]
# band8 = tif_data[:, :, 7]
new_tif_data = np.stack((band1, band2, band3))
print("new_tif_data.shape::", new_tif_data.shape)
new_tif_data = new_tif_data.transpose(1, 2, 0)
print(new_tif_data.shape)
new_tif_data = new_tif_data / (np.max(new_tif_data)) * 255
new_tif_data = new_tif_data.astype(np.uint8)
return new_tif_data
def tif_bands_math2(tif_data):
# 读取image.tif 4bands 转换为uint16 3bands
h, w, bands = tif_data.shape
print("获取函数传递的tif数据为：", tif_data.shape)
band1 = tif_data[:, :, 0]
band2 = tif_data[:, :, 1]
band3 = tif_data[:, :, 2]
band4 = tif_data[:, :, 3]
band5 = tif_data[:, :, 4]
band6 = tif_data[:, :, 5]
# band7 = tif_data[:, :, 6]
# band8 = tif_data[:, :, 7]
new_tif_data = np.stack((band1, band2, band3))
print("new_tif_data.shape::", new_tif_data.shape)
new_tif_data = new_tif_data.transpose(1, 2, 0)
print(new_tif_data.shape)
new_tif_data = new_tif_data / (np.max(new_tif_data)) * 255
new_tif_data = new_tif_data.astype(np.uint8)
return new_tif_data
# a = (band4 - band3) * 1.0
# print(type(a))
# b = ((band4 + band3) + 0.00001)/ 1.0
# print(type(b))
# ndvi = a / b
# print(type(ndvi))
#
# new_ndvi = ndvi / (np.max(ndvi)) * 65536
# new_tif_data2 = np.dstack((new_ndvi, band5, band6))
#
# print("new_tif_data2.shape::", new_tif_data2.shape)
# new_tif_data = np.dstack((band1, band2, band3, band4))
# new_tif_data = np.dstack((band1, band2, band3))
# new_tif_data = new_tif_data / (np.max(new_tif_data)) * 255
# new_tif_data = new_tif_data.astype(np.uint8)
# return new_tif_data
'''
ndvi
h, w = band3.shape
ndvi = np.zeros((h, w)).astype(np.float32)
for i in range(h):
for j in range(w):
# if (band6[i][j] + band3[i][j]) == 0:
if band6[i][j] == 0 and band3[i][j] == 0:
# ndvi[i][j] = 0
continue
else:
a = band6[i][j].astype(np.float32)
b = band3[i][j].astype(np.float32)
# print(ndvi[i][j], band6[i][j], band3[i][j])
# ndvi[i][j] = (band6[i][j] - band3[i][j]).astype(np.float32) / (band6[i][j] + band3[i][j]).astype(np.float32)
ndvi[i][j] = (a - b).astype(np.float32) / (a + b).astype(np.float32)
# print(ndvi[i][j])
print("ndvi最大值", np.max(ndvi))
print("ndvi最小值", np.min(ndvi))
# for i in range(h):
# for j in range(w):
# ndvi[i][j] = ndvi / np.max(ndvi)
# ndvi = ((ndvi - np.min(ndvi)) / (np.max(ndvi) - np.min(ndvi))) * 255
#
# band5 = ((band5 - np.min(band5)) / (np.max(band5) - np.min(band5))) * 255
# band4 = ((band4 - np.min(band4)) / (np.max(band4) - np.min(band4))) * 255
ndvi = ((ndvi - np.min(ndvi)) / (np.max(ndvi) - np.min(ndvi)) * 255).astype(np.uint8)
new_tif_data = np.stack((ndvi, ndvi, ndvi))
# new_tif_data = np.stack((ndvi, band4, band5))
print("new_tif_data.shape::", new_tif_data.shape)
new_tif_data = new_tif_data.transpose(1, 2, 0)
print(new_tif_data.shape)
'''
def single_tif2pngORjpg(tifPath, savePath):
# driver = gdal.GetDriverByName('JPEG')
driver = gdal.GetDriverByName('PNG')
data0 = gdal.Open(tifPath)
# data0.astype(np.byte)
""" ERROR 6: PNG driver doesn't support data type Int16.
# Only eight bit (Byte) and sixteen bit (UInt16) bands supported."""
# data1 = driver.CreateCopy(savePath, data0)
driver.CreateCopy(savePath, data0)
print("路径{}中单个tif数据格式转换完成。。。。。。".format(tifPath))
def tif2pngORjpg(tiffilePath, tifsavePath):
# file_path = r"/home/dsj/dsj_Lab/lmy/Adata/zhuhaiGF/data/clipTIF/imageTIFF/"
# save_path = r"/home/dsj/dsj_Lab/lmy/Adata/zhuhaiGF/data/JPG/"
file_path = tiffilePath
save_path = tifsavePath
driver = gdal.GetDriverByName('PNG')
files = [f for f in os.listdir(file_path) if f.endswith('.tif')]
for each_file in files:
file = file_path + '\\' + each_file # 各tif数据路径全称
fileName, fileLastName = os.path.splitext(each_file) # 文件名， 后缀名
oridata = gdal.Open(file)
driver.CreateCopy(save_path + '\\' + fileName + '.png', oridata)
print("路径{}中所有tif数据格式转换完成。。。。。。".format(file_path))
def single_cutImg(imgPath, savePath):
# 拆分影像图的文件名称
imgFirstName, imgLastName = os.path.splitext(imgPath)
# img = Image.open(imgPath)
# print(type(img))
# print(img.size) 宽，高
img = cv.imread(imgPath, -1)
# print(img.dtype)
# 第二个参数是通道数和位深的参数，
'''
# IMREAD_UNCHANGED = -1 # 不进行转化，比如保存为了16位的图片，读取出来仍然为16位。
# IMREAD_GRAYSCALE = 0 # 进行转化为灰度图，比如保存为了16位的图片，读取出来为8位，类型为CV_8UC1。
# IMREAD_COLOR = 1 # 进行转化为RGB三通道图像，图像深度转为8位
# IMREAD_ANYDEPTH = 2 # 保持图像深度不变，进行转化为灰度图。
# IMREAD_ANYCOLOR = 4 # 若图像通道数小于等于3，则保持原通道数不变；若通道数大于3则只取取前三个通道。图像深度转为8位
# cv.imwrite(ResultPath1 + a + ".png", img) # 保存为png格式
'''
# width, hight = img.size
# print(img.shape)
if len(img.shape) == 2:
imgLabel = Image.open(imgPath)
width, hight = imgLabel.size
elif len(img.shape) == 3:
hight, width, bands = img.shape
# w = 512 # 宽度
# h = 512 # 高度
w = 512
h = 512
_id = 1 # 裁剪结果保存文件名：0 - N 升序方式
y = 0
while y + h <= hight: # 控制高度,图像多余固定尺寸总和部分不要了
x = 0
while x + w <= width: # 控制宽度，图像多余固定尺寸总和部分不要了
# new_img = img.crop((x, y, x + w, y + h))
# new_img.save(ResultPath + a + "_" + str(_id) + b)
# new_img.save(savePath + '\\' + 'img' + "_" + str(_id) + imgLastName)
if len(img.shape) == 2:
new_img = imgLabel.crop((x, y, x + w, y + h))
# new_img.save(ResultPath + a + "_" + str(_id) + b)
# values, counts = np.unique(new_img, return_counts=True)
new_img.save(savePath + '\\' + 'img' + "_" + str(_id) + imgLastName)
# new_img = img[y: y+h, x: x+w]
elif len(img.shape) == 3:
new_img = img[y: y + h, x: x + w, :]
# values, counts = np.unique(new_img, return_counts=True)
cv.imwrite(savePath + '\\' + "img_" + str(_id) + '.png', new_img)
_id += 1
x += w
y = y + h
def cutImg_Slide(imgPath, savePath, image_size, stride):
# 拆分影像图的文件名称
imgFirstName, imgLastName = os.path.splitext(imgPath)
img = cv.imread(imgPath, -1)
if len(img.shape) == 2:
height, width = img.shape
elif len(img.shape) == 3:
height, width, bands = img.shape
_id = 0 # 裁剪结果保存文件名：0 - N 升序方式
h_num = int(height / stride)
w_num = int(width / stride)
print(h_num)
print(w_num)
for i in range(0, h_num):
for j in range(0, w_num):
if i * stride + image_size <= height and j * stride + image_size <= width:
if len(img.shape) == 2:
new_img = img[i * stride: i * stride + image_size, j * stride: j * stride + image_size]
cv.imwrite(savePath + '\\' + "img_" + str(_id) + '.png', new_img)
elif len(img.shape) == 3:
new_img = img[i * stride: i * stride + image_size, j * stride: j * stride + image_size, :]
cv.imwrite(savePath + '\\' + "img_" + str(_id) + '.png', new_img)
_id += 1
def cutImg_Slide_2(file_path, save_path, image_size, stride):
files = [f for f in os.listdir(file_path) if f.endswith('.png')]
_num = 1
for each_file in files:
file = file_path + '\\' + each_file # 各tif数据路径全称
fileName, fileLastName = os.path.splitext(each_file) # 文件名， 后缀名
#oridata = gdal.Open(file)
#driver.CreateCopy(save_path + '\\' + fileName + '.png', oridata)
# 拆分影像图的文件名称
imgFirstName, imgLastName = os.path.splitext(file)
img = cv.imread(file, -1)
if len(img.shape) == 2:
height, width = img.shape
elif len(img.shape) == 3:
height, width, bands = img.shape
_id = 0 # 裁剪结果保存文件名：0 - N 升序方式
h_num = int(height / stride)
w_num = int(width / stride)
print(h_num)
print(w_num)
for i in range(0, h_num):
for j in range(0, w_num):
if i * stride + image_size <= height and j * stride + image_size <= width:
if len(img.shape) == 2:
new_img = img[i * stride: i * stride + image_size, j * stride: j * stride + image_size]
cv.imwrite(save_path + '\\' + "img_" +str(_num) +"_img_" + str(_id) + '.png', new_img)
elif len(img.shape) == 3:
new_img = img[i * stride: i * stride + image_size, j * stride: j * stride + image_size, :]
cv.imwrite(save_path + '\\' +"img_" +str(_num) +"_img_" + str(_id) +'.png', new_img)
_id += 1
_num+=1
#
# # 拆分影像图的文件名称
# imgFirstName, imgLastName = os.path.splitext(imgPath)
# img = cv.imread(imgPath, -1)
#
# if len(img.shape) == 2:
# height, width = img.shape
# elif len(img.shape) == 3:
# height, width, bands = img.shape
#
# _id = 0 # 裁剪结果保存文件名：0 - N 升序方式
# h_num = int(height / stride)
# w_num = int(width / stride)
# print(h_num)
# print(w_num)
#
# for i in range(0, h_num):
# for j in range(0, w_num):
# if i * stride + image_size <= height and j * stride + image_size <= width:
# if len(img.shape) == 2:
# new_img = img[i * stride: i * stride + image_size, j * stride: j * stride + image_size]
# cv.imwrite(savePath + '\\' + "img_" + str(_id) + '.png', new_img)
# elif len(img.shape) == 3:
# new_img = img[i * stride: i * stride + image_size, j * stride: j * stride + image_size, :]
# cv.imwrite(savePath + '\\' + "img_" + str(_id) + '.png', new_img)
# _id += 1
def makeTIFF(tif_dir, tif_saveDir):
tif_data, tif_proj, tif_geotrans = read_gdal(tif_dir)
print("读取的tif数据为：", tif_data.shape)
new_tif_data = tif_bands_math(tif_data)
write_gdal(new_tif_data, tif_saveDir, tif_proj, tif_geotrans)
if __name__ == '__main__':
# path = r'E:\New_gj\dataset\T0\data_pre\labels\label_1.tif'
# img, _, _ = read_gdal(path)
#
# values, counts = np.unique(img, return_counts=True)
# print(values)
# print(counts)
# exit()
path = r'E:\New_gj\dataset\T_multi_22\data_pre' #图像路径
dataset_path = r'E:\New_gj\dataset\T_multi_22\cut' #存储路径
# img
#读取tif bands 转换为3bands
#tif_dir = r'E:\New_gj\dataset\T2\data_pre\images\image.tif'
#tif_saveDir = path + r'\images\image.tif' #三波段图片名
#makeTIFF(tif_dir, tif_saveDir) #转三波段
tif_saveDir=path + r'\images'
pngORjpg_path = path + r'\images\png' # png，jpg存储路径
# tif转格式 路径不要出现中文
#pngORjpg_path = path + r'\images\image.png' #png，jpg存储路径
#single_tif2pngORjpg(tif_saveDir, pngORjpg_path)
tif2pngORjpg(tif_saveDir, pngORjpg_path)
# 裁剪
image_size = 512#输出尺寸
stride = 512
pngORjpg_savePath = dataset_path + r'\images_pre' #保存路径
cutImg_Slide_2(pngORjpg_path, pngORjpg_savePath, image_size, stride)
# -------------------------------------------------------------------------------
# label
tif_saveDir_label = path + r'\labels'
pngORjpg_path_label = path + r'\labels\png'
#single_tif2pngORjpg(tif_saveDir_label, pngORjpg_path_label)
tif2pngORjpg(tif_saveDir_label, pngORjpg_path_label)
pngORjpg_savePath_label = dataset_path + r'\labels_pre'
cutImg_Slide_2(pngORjpg_path_label, pngORjpg_savePath_label, image_size, stride)单张图片裁剪
import os
# import gdal_makeData
import numpy as np
from osgeo import gdal
os.environ["OPENCV_IO_MAX_IMAGE_PIXELS"] = pow(2,40).__str__()
import cv2 as cv
from PIL import Image
from PIL import ImageFile
ImageFile.LOAD_TRUNCATED_IMAGES = True
Image.MAX_IMAGE_PIXELS = None
'''
single_cutImg函数的imread可以将输入的png格式的图片转为8位彩色后，再进行裁剪
'''
def read_gdal(path):
"""
读取一个tiff图像
:param path: 要读取的图像路径(包括后缀名)
:type path: string
:return im_data: 返回图像矩阵(h, w, c)
:rtype im_data: numpy
:return im_proj: 返回投影信息
:rtype im_proj: ?
:return im_geotrans: 返回坐标信息
:rtype im_geotrans: ?
"""
image = gdal.Open(path) # 打开该图像
if image is None:
print(path + "文件无法打开")
return
img_w = image.RasterXSize # 栅格矩阵的列数
img_h = image.RasterYSize # 栅格矩阵的行数
im_bands = image.RasterCount # 波段数
im_proj = image.GetProjection() # 获取投影信息
im_geotrans = image.GetGeoTransform() # 仿射矩阵
im_data = image.ReadAsArray(0, 0, img_w, img_h)
# 二值图一般是二维，需要添加一个维度
if len(im_data.shape) == 2:
im_data = im_data[np.newaxis, :, :]
im_data = im_data.transpose((1, 2, 0))
print("栅格矩阵的列数: ", img_w)
print("栅格矩阵的行数: ", img_h)
print("栅格矩阵的波段数: ", im_bands)
print("栅格矩阵的投影信息: ", im_proj)
print("栅格矩阵的仿射矩阵信息: ", im_geotrans)
print("ia_data形状：", im_data.shape)
return im_data, im_proj, im_geotrans
def write_gdal(im_data, path, im_proj=None, im_geotrans=None):
"""
重新写一个tiff图像
:param im_data: 图像矩阵(h, w, c)
:type im_data: numpy
:param im_proj: 要设置的投影信息(默认None)
:type im_proj: ?
:param im_geotrans: 要设置的坐标信息(默认None)
:type im_geotrans: ?
:param path: 生成的图像路径(包括后缀名)
:type path: string
:return: None
:rtype: None
"""
im_data = im_data.transpose((2, 0, 1))
if 'int8' in im_data.dtype.name:
datatype = gdal.GDT_Byte
elif 'int16' in im_data.dtype.name:
datatype = gdal.GDT_UInt16
elif 'float32' in im_data.dtype.name:
datatype = gdal.GDT_Float32
else:
datatype = gdal.GDT_Float64
if len(im_data.shape) == 3:
im_bands, im_height, im_width = im_data.shape
elif len(im_data.shape) == 2:
im_data = np.array([im_data])
else:
im_bands, (im_height, im_width) = 1, im_data.shape
# 创建文件
driver = gdal.GetDriverByName("GTiff")
dataset = driver.Create(path, im_width, im_height, im_bands, datatype)
if dataset is not None:
if im_geotrans is None or im_proj is None:
pass
else:
dataset.SetGeoTransform(im_geotrans) # 写入仿射变换参数
dataset.SetProjection(im_proj) # 写入投影
for i in range(im_bands):
dataset.GetRasterBand(i + 1).WriteArray(im_data[i])
del dataset
def single_set_proj_trans(ori_path, target_path):
"""
为 target_path 影像设置 ori_path 影像的投影、坐标信息
:param ori_path: 获取 ori_path 影像路径
:type ori_path: string
:param target_path: 获取 target_path 影像路径
:type target_path: string
:return: None
:rtype: None
"""
# 原图像导入
_, im_proj, im_geotrans = read_gdal(ori_path)
# 目标二值图导入
im_data, _, _ = read_gdal(target_path)
# 地理信息写入
write_gdal(im_data, target_path, im_proj, im_geotrans)
def tif_bands_math(tif_data):
# 读取image.tif 4bands 转换为uint16 3bands
h, w, bands = tif_data.shape
print("获取函数传递的tif数据为：", tif_data.shape)
band1 = tif_data[:, :, 0]
band2 = tif_data[:, :, 1]
band3 = tif_data[:, :, 2]
# band4 = tif_data[:, :, 3]
# band5 = tif_data[:, :, 4]
# band6 = tif_data[:, :, 5]
# band7 = tif_data[:, :, 6]
# band8 = tif_data[:, :, 7]
new_tif_data = np.stack((band1, band2, band3))
print("new_tif_data.shape::", new_tif_data.shape)
new_tif_data = new_tif_data.transpose(1, 2, 0)
print(new_tif_data.shape)
new_tif_data = new_tif_data / (np.max(new_tif_data)) * 255
new_tif_data = new_tif_data.astype(np.uint8)
return new_tif_data
def tif_bands_math2(tif_data):
# 读取image.tif 4bands 转换为uint16 3bands
h, w, bands = tif_data.shape
print("获取函数传递的tif数据为：", tif_data.shape)
band1 = tif_data[:, :, 0]
band2 = tif_data[:, :, 1]
band3 = tif_data[:, :, 2]
band4 = tif_data[:, :, 3]
band5 = tif_data[:, :, 4]
band6 = tif_data[:, :, 5]
# band7 = tif_data[:, :, 6]
# band8 = tif_data[:, :, 7]
new_tif_data = np.stack((band1, band2, band3))
print("new_tif_data.shape::", new_tif_data.shape)
new_tif_data = new_tif_data.transpose(1, 2, 0)
print(new_tif_data.shape)
new_tif_data = new_tif_data / (np.max(new_tif_data)) * 255
new_tif_data = new_tif_data.astype(np.uint8)
return new_tif_data
# a = (band4 - band3) * 1.0
# print(type(a))
# b = ((band4 + band3) + 0.00001)/ 1.0
# print(type(b))
# ndvi = a / b
# print(type(ndvi))
#
# new_ndvi = ndvi / (np.max(ndvi)) * 65536
# new_tif_data2 = np.dstack((new_ndvi, band5, band6))
#
# print("new_tif_data2.shape::", new_tif_data2.shape)
# new_tif_data = np.dstack((band1, band2, band3, band4))
# new_tif_data = np.dstack((band1, band2, band3))
# new_tif_data = new_tif_data / (np.max(new_tif_data)) * 255
# new_tif_data = new_tif_data.astype(np.uint8)
# return new_tif_data
'''
ndvi
h, w = band3.shape
ndvi = np.zeros((h, w)).astype(np.float32)
for i in range(h):
for j in range(w):
# if (band6[i][j] + band3[i][j]) == 0:
if band6[i][j] == 0 and band3[i][j] == 0:
# ndvi[i][j] = 0
continue
else:
a = band6[i][j].astype(np.float32)
b = band3[i][j].astype(np.float32)
# print(ndvi[i][j], band6[i][j], band3[i][j])
# ndvi[i][j] = (band6[i][j] - band3[i][j]).astype(np.float32) / (band6[i][j] + band3[i][j]).astype(np.float32)
ndvi[i][j] = (a - b).astype(np.float32) / (a + b).astype(np.float32)
# print(ndvi[i][j])
print("ndvi最大值", np.max(ndvi))
print("ndvi最小值", np.min(ndvi))
# for i in range(h):
# for j in range(w):
# ndvi[i][j] = ndvi / np.max(ndvi)
# ndvi = ((ndvi - np.min(ndvi)) / (np.max(ndvi) - np.min(ndvi))) * 255
#
# band5 = ((band5 - np.min(band5)) / (np.max(band5) - np.min(band5))) * 255
# band4 = ((band4 - np.min(band4)) / (np.max(band4) - np.min(band4))) * 255
ndvi = ((ndvi - np.min(ndvi)) / (np.max(ndvi) - np.min(ndvi)) * 255).astype(np.uint8)
new_tif_data = np.stack((ndvi, ndvi, ndvi))
# new_tif_data = np.stack((ndvi, band4, band5))
print("new_tif_data.shape::", new_tif_data.shape)
new_tif_data = new_tif_data.transpose(1, 2, 0)
print(new_tif_data.shape)
'''
def single_tif2pngORjpg(tifPath, savePath):
# driver = gdal.GetDriverByName('JPEG')
driver = gdal.GetDriverByName('PNG')
data0 = gdal.Open(tifPath)
# data0.astype(np.byte)
""" ERROR 6: PNG driver doesn't support data type Int16.
# Only eight bit (Byte) and sixteen bit (UInt16) bands supported."""
# data1 = driver.CreateCopy(savePath, data0)
driver.CreateCopy(savePath, data0)
print("路径{}中单个tif数据格式转换完成。。。。。。".format(tifPath))
def tif2pngORjpg(tiffilePath, tifsavePath):
# file_path = r"/home/dsj/dsj_Lab/lmy/Adata/zhuhaiGF/data/clipTIF/imageTIFF/"
# save_path = r"/home/dsj/dsj_Lab/lmy/Adata/zhuhaiGF/data/JPG/"
file_path = tiffilePath
save_path = tifsavePath
driver = gdal.GetDriverByName('PNG')
files = [f for f in os.listdir(file_path) if f.endswith('.tif')]
for each_file in files:
file = file_path + each_file # 各tif数据路径全称
fileName, fileLastName = os.path.splitext(each_file) # 文件名， 后缀名
oridata = gdal.Open(file)
data = driver.CreateCopy(save_path + 'zhuhai_' + fileName + '.png', oridata)
print("路径{}中所有tif数据格式转换完成。。。。。。".format(file_path))
def single_cutImg(imgPath, savePath):
# 拆分影像图的文件名称
imgFirstName, imgLastName = os.path.splitext(imgPath)
# img = Image.open(imgPath)
# print(type(img))
# print(img.size) 宽，高
img = cv.imread(imgPath, -1)
# print(img.dtype)
# 第二个参数是通道数和位深的参数，
'''
# IMREAD_UNCHANGED = -1 # 不进行转化，比如保存为了16位的图片，读取出来仍然为16位。
# IMREAD_GRAYSCALE = 0 # 进行转化为灰度图，比如保存为了16位的图片，读取出来为8位，类型为CV_8UC1。
# IMREAD_COLOR = 1 # 进行转化为RGB三通道图像，图像深度转为8位
# IMREAD_ANYDEPTH = 2 # 保持图像深度不变，进行转化为灰度图。
# IMREAD_ANYCOLOR = 4 # 若图像通道数小于等于3，则保持原通道数不变；若通道数大于3则只取取前三个通道。图像深度转为8位
# cv.imwrite(ResultPath1 + a + ".png", img) # 保存为png格式
'''
# width, hight = img.size
# print(img.shape)
if len(img.shape) == 2:
imgLabel = Image.open(imgPath)
width, hight = imgLabel.size
elif len(img.shape) == 3:
hight, width, bands = img.shape
# w = 512 # 宽度
# h = 512 # 高度
w = 512
h = 512
_id = 1 # 裁剪结果保存文件名：0 - N 升序方式
y = 0
while y + h <= hight: # 控制高度,图像多余固定尺寸总和部分不要了
x = 0
while x + w <= width: # 控制宽度，图像多余固定尺寸总和部分不要了
# new_img = img.crop((x, y, x + w, y + h))
# new_img.save(ResultPath + a + "_" + str(_id) + b)
# new_img.save(savePath + '\\' + 'img' + "_" + str(_id) + imgLastName)
if len(img.shape) == 2:
new_img = imgLabel.crop((x, y, x + w, y + h))
# new_img.save(ResultPath + a + "_" + str(_id) + b)
# values, counts = np.unique(new_img, return_counts=True)
new_img.save(savePath + '\\' + 'img' + "_" + str(_id) + imgLastName)
# new_img = img[y: y+h, x: x+w]
elif len(img.shape) == 3:
new_img = img[y: y + h, x: x + w, :]
# values, counts = np.unique(new_img, return_counts=True)
cv.imwrite(savePath + '\\' + "img_" + str(_id) + '.png', new_img)
_id += 1
x += w
y = y + h
def cutImg_Slide(imgPath, savePath, image_size, stride):
# 拆分影像图的文件名称
imgFirstName, imgLastName = os.path.splitext(imgPath)
img = cv.imread(imgPath, -1)
if len(img.shape) == 2:
height, width = img.shape
elif len(img.shape) == 3:
height, width, bands = img.shape
_id = 0 # 裁剪结果保存文件名：0 - N 升序方式
h_num = int(height / stride)
w_num = int(width / stride)
print(h_num)
print(w_num)
for i in range(0, h_num):
for j in range(0, w_num):
if i * stride + image_size <= height and j * stride + image_size <= width:
if len(img.shape) == 2:
new_img = img[i * stride: i * stride + image_size, j * stride: j * stride + image_size]
cv.imwrite(savePath + '\\' + imgFirstName + "_img_" + str(_id) + '.png', new_img)
elif len(img.shape) == 3:
new_img = img[i * stride: i * stride + image_size, j * stride: j * stride + image_size, :]
cv.imwrite(savePath + '\\' + imgFirstName +"_img_" + str(_id) + '.png', new_img)
_id += 1
def makeTIFF(tif_dir, tif_saveDir):
tif_data, tif_proj, tif_geotrans = read_gdal(tif_dir)
print("读取的tif数据为：", tif_data.shape)
new_tif_data = tif_bands_math(tif_data)
write_gdal(new_tif_data, tif_saveDir, tif_proj, tif_geotrans)
if __name__ == '__main__':
# path = r'E:\New_gj\dataset\T0\data_pre\labels\label_1.tif'
# img, _, _ = read_gdal(path)
#
# values, counts = np.unique(img, return_counts=True)
# print(values)
# print(counts)
# exit()
path = r'E:\New_gj\dataset\T_multi_re_0.1\predict' #图像路径
dataset_path = r'E:\New_gj\dataset\T_multi_re_0.1\predict\cut' #存储路径
# img
#读取tif bands 转换为3bands
#tif_dir = r'E:\New_gj\dataset\T2\data_pre\images\image.tif'
tif_saveDir = path + r'\images\image_2.tif' #三波段图片名
#makeTIFF(tif_dir, tif_saveDir) #转三波段
# tif转格式 路径不要出现中文
pngORjpg_path = path + r'\images\png\image_2.png' #png，jpg存储路径
single_tif2pngORjpg(tif_saveDir, pngORjpg_path)
# 裁剪
image_size = 512#输出尺寸
stride = 512
pngORjpg_savePath = dataset_path + r'\images' #保存路径
cutImg_Slide(pngORjpg_path, pngORjpg_savePath, image_size, stride)
# -------------------------------------------------------------------------------
# label
tif_saveDir_label = path + r'\labels\label_2.tif'
pngORjpg_path_label = path + r'\labels\png\label_2.png'
single_tif2pngORjpg(tif_saveDir_label, pngORjpg_path_label)
pngORjpg_savePath_label = dataset_path + r'\labels'
cutImg_Slide(pngORjpg_path_label, pngORjpg_savePath_label, image_size, stride)本文章为转载内容，我们尊重原作者对文章享有的著作权。如有内容错误或侵权问题，欢迎原作者联系我们进行内容更正或删除文章。
