389 lines
15 KiB
Python
389 lines
15 KiB
Python
|
|
|||
|
|
|
|||
|
|
import os
|
|||
|
|
import cv2
|
|||
|
|
import numpy as np
|
|||
|
|
import rasterio
|
|||
|
|
import geopandas as gpd
|
|||
|
|
from shapely.geometry import Polygon
|
|||
|
|
from rasterio.features import shapes
|
|||
|
|
from scipy.ndimage import binary_opening, binary_closing, binary_fill_holes, label
|
|||
|
|
from tqdm import tqdm
|
|||
|
|
|
|||
|
|
# class PredictionVisualizer:
|
|||
|
|
# def __init__(self, image_path, pred_mask_path):
|
|||
|
|
# self.image_path = image_path
|
|||
|
|
# self.pred_mask_path = pred_mask_path
|
|||
|
|
# # 检查输入图像的地理参考信息
|
|||
|
|
# with rasterio.open(image_path) as src:
|
|||
|
|
# self.has_georeference = src.transform.is_identity == False
|
|||
|
|
# if not self.has_georeference:
|
|||
|
|
# print("警告：输入图像没有地理参考信息，将使用像素坐标系统。")
|
|||
|
|
# self.transform = src.transform
|
|||
|
|
# self.crs = src.crs
|
|||
|
|
# self.width = src.width
|
|||
|
|
# self.height = src.height
|
|||
|
|
|
|||
|
|
# def post_process_mask(self, mask, min_area=100, kernel_size=(5,5)):
|
|||
|
|
# mask = mask.astype(np.uint8)
|
|||
|
|
# kernel = np.ones(kernel_size)
|
|||
|
|
# mask = binary_closing(mask, structure=kernel)
|
|||
|
|
# mask = binary_opening(mask, structure=np.ones((3, 3)))
|
|||
|
|
# mask = binary_fill_holes(mask)
|
|||
|
|
# labeled_array, num_features = label(mask)
|
|||
|
|
# for i in range(1, num_features + 1):
|
|||
|
|
# component = (labeled_array == i)
|
|||
|
|
# if np.sum(component) < min_area:
|
|||
|
|
# mask[component] = 0
|
|||
|
|
# return mask.astype(np.uint8)
|
|||
|
|
|
|||
|
|
# def draw_contours_on_image(self, save_path=None, color=(0, 0, 255), thickness=2, approx_method=cv2.CHAIN_APPROX_TC89_KCOS):
|
|||
|
|
# with rasterio.open(self.image_path) as src:
|
|||
|
|
# image = src.read([1, 2, 3]).transpose(1, 2, 0)
|
|||
|
|
# image = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)
|
|||
|
|
|
|||
|
|
# pred_mask = cv2.imread(self.pred_mask_path, cv2.IMREAD_GRAYSCALE)
|
|||
|
|
# _, binary = cv2.threshold(pred_mask, 127, 255, cv2.THRESH_BINARY)
|
|||
|
|
# processed_mask = self.post_process_mask(binary, min_area=10)
|
|||
|
|
|
|||
|
|
# contours, _ = cv2.findContours(processed_mask, cv2.RETR_EXTERNAL, approx_method)
|
|||
|
|
|
|||
|
|
# simplified_contours = []
|
|||
|
|
# for cnt in contours:
|
|||
|
|
# if len(cnt) >= 3:
|
|||
|
|
# epsilon = 0.001 * cv2.arcLength(cnt, True) if self.has_georeference else 0.002 * cv2.arcLength(cnt, True)
|
|||
|
|
# approx = cv2.approxPolyDP(cnt, epsilon, True)
|
|||
|
|
# if len(approx) >= 3:
|
|||
|
|
# simplified_contours.append(approx.astype(np.int32))
|
|||
|
|
|
|||
|
|
# result = image.copy()
|
|||
|
|
# cv2.drawContours(result, simplified_contours, -1, color, thickness)
|
|||
|
|
# if save_path:
|
|||
|
|
# cv2.imwrite(save_path, result)
|
|||
|
|
|
|||
|
|
# return result, simplified_contours
|
|||
|
|
|
|||
|
|
# def export_contours_to_shapefile(self, contours, output_path):
|
|||
|
|
# if not self.has_georeference:
|
|||
|
|
# print("当前图像没有地理参考信息，无法导出地理坐标。")
|
|||
|
|
# return None
|
|||
|
|
|
|||
|
|
# polygons = []
|
|||
|
|
# for cnt in contours:
|
|||
|
|
# coords = []
|
|||
|
|
# for point in cnt[:, 0, :]:
|
|||
|
|
# x_pixel, y_pixel = point[0], point[1]
|
|||
|
|
# x_geo, y_geo = rasterio.transform.xy(self.transform, y_pixel, x_pixel)
|
|||
|
|
# coords.append((x_geo, y_geo))
|
|||
|
|
# if coords[0] != coords[-1]:
|
|||
|
|
# coords.append(coords[0])
|
|||
|
|
# poly = Polygon(coords)
|
|||
|
|
# if poly.is_valid:
|
|||
|
|
# polygons.append(poly)
|
|||
|
|
|
|||
|
|
# if not polygons:
|
|||
|
|
# print(f"未生成有效地理轮廓：{output_path}")
|
|||
|
|
# return None
|
|||
|
|
|
|||
|
|
# gdf = gpd.GeoDataFrame({'geometry': polygons}, crs=self.crs)
|
|||
|
|
# gdf.to_file(output_path)
|
|||
|
|
# print(f"✅ 已保存地理红框到 {output_path}")
|
|||
|
|
# return output_path
|
|||
|
|
|
|||
|
|
|
|||
|
|
# def process_all_images(image1_dir, image2_dir, pred_mask_dir, save_dir, shp_dir):
|
|||
|
|
# os.makedirs(save_dir, exist_ok=True)
|
|||
|
|
# os.makedirs(shp_dir, exist_ok=True)
|
|||
|
|
|
|||
|
|
# mask_files = sorted([f for f in os.listdir(pred_mask_dir) if f.endswith('.png')])
|
|||
|
|
|
|||
|
|
# for mask_file in tqdm(mask_files, desc="处理影像中"):
|
|||
|
|
# base_name = os.path.splitext(mask_file)[0]
|
|||
|
|
# image1_path = os.path.join(image1_dir, base_name + '.png')
|
|||
|
|
# image2_path = os.path.join(image2_dir, base_name + '.png')
|
|||
|
|
# mask_path = os.path.join(pred_mask_dir, mask_file)
|
|||
|
|
# save_path_img1 = os.path.join(save_dir, f'{base_name}_image1_red.png')
|
|||
|
|
# save_path_img2 = os.path.join(save_dir, f'{base_name}_image2_red.png')
|
|||
|
|
# shp_path = os.path.join(shp_dir, f'{base_name}.shp')
|
|||
|
|
|
|||
|
|
# if not os.path.exists(image1_path) or not os.path.exists(image2_path):
|
|||
|
|
# print(f"[跳过] 缺失影像：{base_name}")
|
|||
|
|
# continue
|
|||
|
|
|
|||
|
|
# # 1. 生成轮廓 + 导出shp（基于image1）
|
|||
|
|
# visualizer = PredictionVisualizer(image1_path, mask_path)
|
|||
|
|
# _, contours = visualizer.draw_contours_on_image(save_path=save_path_img1)
|
|||
|
|
# visualizer.export_contours_to_shapefile(contours, shp_path)
|
|||
|
|
|
|||
|
|
# # 2. image2 也绘制红框（但不导出地理坐标）
|
|||
|
|
# visualizer2 = PredictionVisualizer(image2_path, mask_path)
|
|||
|
|
# visualizer2.draw_contours_on_image(save_path=save_path_img2)
|
|||
|
|
|
|||
|
|
|
|||
|
|
# # === 设置路径并运行 ===
|
|||
|
|
# if __name__ == '__main__':
|
|||
|
|
# image1_dir = r'/media/data0/HL/2025-7-6/val/image1'
|
|||
|
|
# image2_dir = r'/media/data0/HL/2025-7-6/val/image2'
|
|||
|
|
# pred_mask_dir = r'/media/data0/HL/CropLand-CD-main_3/CropLand-CD-main/save'
|
|||
|
|
# save_dir = r'/media/data0/HL/CropLand-CD-main_3/CropLand-CD-main/save_red'
|
|||
|
|
# shp_dir = r'/media/data0/HL/CropLand-CD-main_3/CropLand-CD-main/save_shp'
|
|||
|
|
|
|||
|
|
# process_all_images(image1_dir, image2_dir, pred_mask_dir, save_dir, shp_dir)
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
class PredictionVisualizer:
|
|||
|
|
def __init__(self, image_path, pred_mask_path):
|
|||
|
|
self.image_path = image_path
|
|||
|
|
self.pred_mask_path = pred_mask_path
|
|||
|
|
|
|||
|
|
with rasterio.open(image_path) as src:
|
|||
|
|
self.transform = src.transform
|
|||
|
|
self.crs = src.crs
|
|||
|
|
self.width = src.width
|
|||
|
|
self.height = src.height
|
|||
|
|
self.has_georeference = self.crs is not None and self.transform != rasterio.Affine.identity()
|
|||
|
|
|
|||
|
|
# if not self.has_georeference:
|
|||
|
|
# print(f"警告：输入图像 {image_path} 没有地理参考信息，将使用像素坐标系统。")
|
|||
|
|
|
|||
|
|
def post_process_mask(self, mask, min_area=100, kernel_size=(5, 5)):
|
|||
|
|
mask = mask.astype(np.uint8)
|
|||
|
|
kernel = np.ones(kernel_size)
|
|||
|
|
mask = binary_closing(mask, structure=kernel)
|
|||
|
|
mask = binary_opening(mask, structure=np.ones((3, 3)))
|
|||
|
|
mask = binary_fill_holes(mask)
|
|||
|
|
|
|||
|
|
labeled_array, num_features = label(mask)
|
|||
|
|
for i in range(1, num_features + 1):
|
|||
|
|
component = (labeled_array == i)
|
|||
|
|
if np.sum(component) < min_area:
|
|||
|
|
mask[component] = 0
|
|||
|
|
return mask.astype(np.uint8)
|
|||
|
|
|
|||
|
|
def draw_contours_on_image(self, save_path=None, color=(0, 0, 255), thickness=2, approx_method=cv2.CHAIN_APPROX_TC89_KCOS):
|
|||
|
|
with rasterio.open(self.image_path) as src:
|
|||
|
|
image = src.read([1, 2, 3]).transpose(1, 2, 0)
|
|||
|
|
image = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)
|
|||
|
|
|
|||
|
|
pred_mask = cv2.imread(self.pred_mask_path, cv2.IMREAD_GRAYSCALE)
|
|||
|
|
if pred_mask is None:
|
|||
|
|
raise FileNotFoundError(f"无法读取掩膜图像: {self.pred_mask_path}")
|
|||
|
|
_, binary = cv2.threshold(pred_mask, 127, 255, cv2.THRESH_BINARY)
|
|||
|
|
processed_mask = self.post_process_mask(binary, min_area=10)
|
|||
|
|
|
|||
|
|
contours, _ = cv2.findContours(processed_mask, cv2.RETR_EXTERNAL, approx_method)
|
|||
|
|
|
|||
|
|
simplified_contours = []
|
|||
|
|
for cnt in contours:
|
|||
|
|
if len(cnt) >= 3:
|
|||
|
|
epsilon = 0.001 * cv2.arcLength(cnt, True) if self.has_georeference else 0.002 * cv2.arcLength(cnt, True)
|
|||
|
|
approx = cv2.approxPolyDP(cnt, epsilon, True)
|
|||
|
|
if len(approx) >= 3:
|
|||
|
|
simplified_contours.append(approx.astype(np.int32))
|
|||
|
|
|
|||
|
|
result = image.copy()
|
|||
|
|
cv2.drawContours(result, simplified_contours, -1, color, thickness)
|
|||
|
|
|
|||
|
|
if save_path:
|
|||
|
|
cv2.imwrite(save_path, result)
|
|||
|
|
|
|||
|
|
return result, simplified_contours
|
|||
|
|
|
|||
|
|
|
|||
|
|
class ChangeVisualizer:
|
|||
|
|
def __init__(self, image1_dir, image2_dir, pred_mask_dir, save_dir):
|
|||
|
|
self.image1_dir = image1_dir
|
|||
|
|
self.image2_dir = image2_dir
|
|||
|
|
self.pred_mask_dir = pred_mask_dir
|
|||
|
|
self.save_dir = save_dir
|
|||
|
|
os.makedirs(save_dir, exist_ok=True)
|
|||
|
|
|
|||
|
|
@staticmethod
|
|||
|
|
def find_corresponding_file(base_dir, base_name):
|
|||
|
|
for ext in ['.png', '.tif', '.jpg', '.jpeg']:
|
|||
|
|
path = os.path.join(base_dir, base_name + ext)
|
|||
|
|
if os.path.exists(path):
|
|||
|
|
return path
|
|||
|
|
return None
|
|||
|
|
|
|||
|
|
def run(self):
|
|||
|
|
mask_files = sorted([
|
|||
|
|
f for f in os.listdir(self.pred_mask_dir)
|
|||
|
|
if os.path.splitext(f)[1].lower() in ['.png', '.tif', '.jpg', '.jpeg']
|
|||
|
|
])
|
|||
|
|
|
|||
|
|
for mask_file in tqdm(mask_files, desc="处理影像中"):
|
|||
|
|
base_name = os.path.splitext(mask_file)[0]
|
|||
|
|
|
|||
|
|
image1_path = self.find_corresponding_file(self.image1_dir, base_name)
|
|||
|
|
image2_path = self.find_corresponding_file(self.image2_dir, base_name)
|
|||
|
|
mask_path = os.path.join(self.pred_mask_dir, mask_file)
|
|||
|
|
|
|||
|
|
if not image1_path or not image2_path:
|
|||
|
|
print(f"[跳过] 缺失影像：{base_name}")
|
|||
|
|
continue
|
|||
|
|
|
|||
|
|
save_path_img1 = os.path.join(self.save_dir, f'{base_name}_image1_red.png')
|
|||
|
|
save_path_img2 = os.path.join(self.save_dir, f'{base_name}_image2_red.png')
|
|||
|
|
|
|||
|
|
# 1. 从 image1 获取轮廓
|
|||
|
|
visualizer = PredictionVisualizer(image1_path, mask_path)
|
|||
|
|
_, contours = visualizer.draw_contours_on_image(save_path=None)
|
|||
|
|
|
|||
|
|
# 2. 两期图像绘制红框
|
|||
|
|
for img_path, out_path in [(image1_path, save_path_img1), (image2_path, save_path_img2)]:
|
|||
|
|
visualizer_i = PredictionVisualizer(img_path, mask_path)
|
|||
|
|
visualizer_i.draw_contours_on_image(save_path=out_path)
|
|||
|
|
|
|||
|
|
|
|||
|
|
# # 主函数调用方式
|
|||
|
|
# if __name__ == '__main__':
|
|||
|
|
# image1_dir = r'/media/data0/HL/2025-7-6/val/image1'
|
|||
|
|
# image2_dir = r'/media/data0/HL/2025-7-6/val/image2'
|
|||
|
|
# pred_mask_dir = r'/media/data0/HL/CropLand-CD-main_3/CropLand-CD-main/save'
|
|||
|
|
# save_dir = r'/media/data0/HL/CropLand-CD-main_3/CropLand-CD-main/save_red'
|
|||
|
|
|
|||
|
|
# visualizer = ChangeVisualizer(image1_dir, image2_dir, pred_mask_dir, save_dir)
|
|||
|
|
# visualizer.run()
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
# class PredictionVisualizer:
|
|||
|
|
# def __init__(self, image_path, pred_mask_path):
|
|||
|
|
# self.image_path = image_path
|
|||
|
|
# self.pred_mask_path = pred_mask_path
|
|||
|
|
|
|||
|
|
# with rasterio.open(image_path) as src:
|
|||
|
|
# self.transform = src.transform
|
|||
|
|
# self.crs = src.crs
|
|||
|
|
# self.width = src.width
|
|||
|
|
# self.height = src.height
|
|||
|
|
# self.has_georeference = self.crs is not None and self.transform != rasterio.Affine.identity()
|
|||
|
|
|
|||
|
|
# if not self.has_georeference:
|
|||
|
|
# print(f"警告：输入图像 {image_path} 没有地理参考信息，将使用像素坐标系统。")
|
|||
|
|
|
|||
|
|
# def post_process_mask(self, mask, min_area=100, kernel_size=(5, 5)):
|
|||
|
|
# mask = mask.astype(np.uint8)
|
|||
|
|
# kernel = np.ones(kernel_size)
|
|||
|
|
# mask = binary_closing(mask, structure=kernel)
|
|||
|
|
# mask = binary_opening(mask, structure=np.ones((3, 3)))
|
|||
|
|
# mask = binary_fill_holes(mask)
|
|||
|
|
|
|||
|
|
# labeled_array, num_features = label(mask)
|
|||
|
|
# for i in range(1, num_features + 1):
|
|||
|
|
# component = (labeled_array == i)
|
|||
|
|
# if np.sum(component) < min_area:
|
|||
|
|
# mask[component] = 0
|
|||
|
|
# return mask.astype(np.uint8)
|
|||
|
|
|
|||
|
|
# def draw_contours_on_image(self, save_path=None, color=(0, 0, 255), thickness=2, approx_method=cv2.CHAIN_APPROX_TC89_KCOS):
|
|||
|
|
# with rasterio.open(self.image_path) as src:
|
|||
|
|
# image = src.read([1, 2, 3]).transpose(1, 2, 0)
|
|||
|
|
# image = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)
|
|||
|
|
|
|||
|
|
# pred_mask = cv2.imread(self.pred_mask_path, cv2.IMREAD_GRAYSCALE)
|
|||
|
|
# if pred_mask is None:
|
|||
|
|
# raise FileNotFoundError(f"无法读取掩膜图像: {self.pred_mask_path}")
|
|||
|
|
# _, binary = cv2.threshold(pred_mask, 127, 255, cv2.THRESH_BINARY)
|
|||
|
|
# processed_mask = self.post_process_mask(binary, min_area=10)
|
|||
|
|
|
|||
|
|
# contours, _ = cv2.findContours(processed_mask, cv2.RETR_EXTERNAL, approx_method)
|
|||
|
|
|
|||
|
|
# simplified_contours = []
|
|||
|
|
# for cnt in contours:
|
|||
|
|
# if len(cnt) >= 3:
|
|||
|
|
# epsilon = 0.001 * cv2.arcLength(cnt, True) if self.has_georeference else 0.002 * cv2.arcLength(cnt, True)
|
|||
|
|
# approx = cv2.approxPolyDP(cnt, epsilon, True)
|
|||
|
|
# if len(approx) >= 3:
|
|||
|
|
# simplified_contours.append(approx.astype(np.int32))
|
|||
|
|
|
|||
|
|
# result = image.copy()
|
|||
|
|
# cv2.drawContours(result, simplified_contours, -1, color, thickness)
|
|||
|
|
|
|||
|
|
# if save_path:
|
|||
|
|
# cv2.imwrite(save_path, result)
|
|||
|
|
|
|||
|
|
# return result, simplified_contours
|
|||
|
|
|
|||
|
|
# def find_corresponding_file(base_dir, base_name):
|
|||
|
|
# """在目录下查找与 base_name 同名（不含扩展名）的图像文件（不限于 .png）"""
|
|||
|
|
# for ext in ['.png', '.tif', '.jpg', '.jpeg']:
|
|||
|
|
# path = os.path.join(base_dir, base_name + ext)
|
|||
|
|
# if os.path.exists(path):
|
|||
|
|
# return path
|
|||
|
|
# return None
|
|||
|
|
|
|||
|
|
# def process_all_images(image1_dir, image2_dir, pred_mask_dir, save_dir):
|
|||
|
|
# os.makedirs(save_dir, exist_ok=True)
|
|||
|
|
|
|||
|
|
# mask_files = sorted([f for f in os.listdir(pred_mask_dir) if os.path.splitext(f)[1].lower() in ['.png', '.tif', '.jpg', '.jpeg']])
|
|||
|
|
|
|||
|
|
# for mask_file in tqdm(mask_files, desc="处理影像中"):
|
|||
|
|
# base_name = os.path.splitext(mask_file)[0]
|
|||
|
|
|
|||
|
|
# image1_path = find_corresponding_file(image1_dir, base_name)
|
|||
|
|
# image2_path = find_corresponding_file(image2_dir, base_name)
|
|||
|
|
# mask_path = os.path.join(pred_mask_dir, mask_file)
|
|||
|
|
|
|||
|
|
# save_path_img1 = os.path.join(save_dir, f'{base_name}_image1_red.png')
|
|||
|
|
# save_path_img2 = os.path.join(save_dir, f'{base_name}_image2_red.png')
|
|||
|
|
|
|||
|
|
# if not image1_path or not image2_path:
|
|||
|
|
# print(f"[跳过] 缺失影像：{base_name}")
|
|||
|
|
# continue
|
|||
|
|
|
|||
|
|
# # 先获取轮廓
|
|||
|
|
# visualizer = PredictionVisualizer(image1_path, mask_path)
|
|||
|
|
# _, contours = visualizer.draw_contours_on_image(save_path=None)
|
|||
|
|
|
|||
|
|
# for img_path, out_path in [(image1_path, save_path_img1), (image2_path, save_path_img2)]:
|
|||
|
|
# visualizer_i = PredictionVisualizer(img_path, mask_path)
|
|||
|
|
# visualizer_i.draw_contours_on_image(save_path=out_path)
|
|||
|
|
|
|||
|
|
# # === 设置路径并运行 ===
|
|||
|
|
# if __name__ == '__main__':
|
|||
|
|
# image1_dir = r'/media/data0/HL/2025-7-6/val/image1'
|
|||
|
|
# image2_dir = r'/media/data0/HL/2025-7-6/val/image2'
|
|||
|
|
# pred_mask_dir = r'/media/data0/HL/CropLand-CD-main_3/CropLand-CD-main/save'
|
|||
|
|
# save_dir = r'/media/data0/HL/CropLand-CD-main_3/CropLand-CD-main/save_red'
|
|||
|
|
|
|||
|
|
# process_all_images(image1_dir, image2_dir, pred_mask_dir, save_dir)
|