基于OpenCV的cv2.Canny与cv2.findContours提取异质卫星图像有效区域

Hey there! Welcome to the platform—happy to help you tackle this satellite image processing task. Let’s break down how to locate that largest valid region, simplify its contour to a small set of points (max 15), and save them out. Here’s a practical, step-by-step approach tailored to your needs:

First, you need to define what counts as "valid data"—this usually means excluding black borders, cloud cover, or pixels marked as NoData (common in satellite imagery, often 0, NaN, or a specific value like 65535 for 16-bit files). We’ll create a binary mask to separate valid vs. invalid pixels:

import cv2
import numpy as np

# Load your satellite image (adjust path/band as needed; use grayscale for simplicity)
img = cv2.imread("satellite_image.tif", cv2.IMREAD_GRAYSCALE)

# Define valid pixel range (tweak this based on your image's actual NoData values)
valid_mask = (img > 0) & (img != 65535)  # Example: exclude 0 and 65535
valid_mask = valid_mask.astype(np.uint8) * 255  # Convert to 8-bit mask for OpenCV compatibility

Next, we’ll detect all connected valid regions and pick the one with the largest area (since you want the biggest valid data region):

# Find outer contours only (avoids nested small regions inside larger ones)
contours, _ = cv2.findContours(valid_mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)

# Sort contours by area and grab the largest one
largest_contour = max(contours, key=cv2.contourArea)

Use the Douglas-Peucker algorithm (via OpenCV’s approxPolyDP) to reduce the number of contour points. Adjust the epsilon parameter (a percentage of the contour perimeter) until you get a point count ≤15:

# Calculate the contour's perimeter
perimeter = cv2.arcLength(largest_contour, closed=True)

# Start with a small epsilon (2% of perimeter) and adjust incrementally
epsilon = 0.02 * perimeter
simplified_contour = cv2.approxPolyDP(largest_contour, epsilon, closed=True)

# If we still have too many points, increase epsilon until we hit the 15-point limit
while len(simplified_contour) > 15:
    epsilon += 0.005 * perimeter
    simplified_contour = cv2.approxPolyDP(largest_contour, epsilon, closed=True)

# Convert contour points to clean (x,y) tuples
simplified_points = [tuple(point[0]) for point in simplified_contour]
print(f"Simplified contour to {len(simplified_points)} points: {simplified_points}")

Save the simplified points to a format that works for your workflow—text file, JSON, etc.:

# Save to a plain text file (one point per line)
with open("contour_points.txt", "w") as f:
    for x, y in simplified_points:
        f.write(f"{x}, {y}\n")

# Or save as JSON for easier parsing later
import json
with open("contour_points.json", "w") as f:
    json.dump(simplified_points, f)

• Adjust Valid Mask Logic: Since your images are diverse, you may need to tweak the valid pixel check for different image types (e.g., multi-band imagery might require all relevant bands to be non-NoData).
• Batch Processing: Wrap this code in a loop using os.listdir or pathlib to process hundreds of images automatically.
• Edge Cases: For perfect square regions, approxPolyDP will automatically reduce the contour to 4 points—exactly what you need for simple scenarios.

内容的提问来源于stack exchange，提问作者rudypl2