Change Detection with Instance Segmentation¶

This notebook demonstrates the change detection functionality in GeoAI, which provides instance segmentation and confidence scoring for individual change objects.

The change detection functionality builds upon the torchange package developed by Dr. Zhuo Zheng. We have made it much easier to analyze remote sensing imagery and visualize the results.

Overview¶

The change detection system provides:

• Instance Segmentation: Each change object gets a unique ID
• Confidence Scores: Individual confidence values for each detected instance
• Proper GeoTIFF Output: Maintains spatial reference information

Key Features¶

• Instance-level change detection with unique IDs
• Confidence scoring for quality assessment
• Support for large GeoTIFF files
• Comprehensive analysis capabilities

Install packages¶

In [ ]:

# %pip install geoai-py

# %pip install geoai-py

Import libraries¶

In [ ]:

import geoai
import os
import numpy as np
import matplotlib.pyplot as plt
from pathlib import Path

import geoai import os import numpy as np import matplotlib.pyplot as plt from pathlib import Path

Setup¶

In [ ]:

# Check if CUDA is available
device = geoai.get_device()
print(f"Using device: {device}")

# Set up paths
out_folder = "change_detection_results"
Path(out_folder).mkdir(exist_ok=True)

print(f"Working directory: {out_folder}")

# Check if CUDA is available device = geoai.get_device() print(f"Using device: {device}") # Set up paths out_folder = "change_detection_results" Path(out_folder).mkdir(exist_ok=True) print(f"Working directory: {out_folder}")

Download sample data¶

We'll use NAIP imagery for Las Vegas to demonstrate change detection.

In [ ]:

# Download NAIP imagery
naip_2019_url = "https://huggingface.co/datasets/giswqs/geospatial/resolve/main/las_vegas_naip_2019_a.tif"
naip_2022_url = "https://huggingface.co/datasets/giswqs/geospatial/resolve/main/las_vegas_naip_2022_a.tif"

naip_2019_path = geoai.download_file(naip_2019_url)
naip_2022_path = geoai.download_file(naip_2022_url)

print(f"Downloaded 2019 NAIP: {naip_2019_path}")
print(f"Downloaded 2022 NAIP: {naip_2022_path}")

# Download NAIP imagery naip_2019_url = "https://huggingface.co/datasets/giswqs/geospatial/resolve/main/las_vegas_naip_2019_a.tif" naip_2022_url = "https://huggingface.co/datasets/giswqs/geospatial/resolve/main/las_vegas_naip_2022_a.tif" naip_2019_path = geoai.download_file(naip_2019_url) naip_2022_path = geoai.download_file(naip_2022_url) print(f"Downloaded 2019 NAIP: {naip_2019_path}") print(f"Downloaded 2022 NAIP: {naip_2022_path}")

Visualize sample data¶

In [ ]:

# Check raster information
geoai.get_raster_info(naip_2019_path)

# Check raster information geoai.get_raster_info(naip_2019_path)

In [ ]:

# View the images
geoai.view_raster(naip_2019_path)

# View the images geoai.view_raster(naip_2019_path)

In [ ]:

geoai.view_raster(naip_2022_path)

geoai.view_raster(naip_2022_path)

Initialize Change Detection¶

Create the change detection system with optimal parameters.

In [ ]:

# Initialize change detection
detector = geoai.ChangeDetection(sam_model_type="vit_h")

# Configure parameters (following the torchange example)
detector.set_hyperparameters(
    change_confidence_threshold=145,
    use_normalized_feature=True,
    bitemporal_match=True,
)

detector.set_mask_generator_params(
    points_per_side=32,
    stability_score_thresh=0.95,
)

print("Change detection system initialized!")

# Initialize change detection detector = geoai.ChangeDetection(sam_model_type="vit_h") # Configure parameters (following the torchange example) detector.set_hyperparameters( change_confidence_threshold=145, use_normalized_feature=True, bitemporal_match=True, ) detector.set_mask_generator_params( points_per_side=32, stability_score_thresh=0.95, ) print("Change detection system initialized!")

Run Change Detection¶

Execute change detection with instance segmentation and confidence scoring.

In [ ]:

# Run change detection
results = detector.detect_changes(
    naip_2019_path,
    naip_2022_path,
    output_path=f"{out_folder}/binary_mask.tif",
    export_probability=True,
    probability_output_path=f"{out_folder}/probability_mask.tif",
    export_instance_masks=True,
    instance_masks_output_path=f"{out_folder}/instance_masks.tif",
    return_detailed_results=True,
    return_results=False,
)

print(f"Change detection completed!")
print(f"Total instances detected: {results['summary']['total_masks']}")
print(f"Image size: {results['summary']['original_shape']}")

# Run change detection results = detector.detect_changes( naip_2019_path, naip_2022_path, output_path=f"{out_folder}/binary_mask.tif", export_probability=True, probability_output_path=f"{out_folder}/probability_mask.tif", export_instance_masks=True, instance_masks_output_path=f"{out_folder}/instance_masks.tif", return_detailed_results=True, return_results=False, ) print(f"Change detection completed!") print(f"Total instances detected: {results['summary']['total_masks']}") print(f"Image size: {results['summary']['original_shape']}")

Analyze Results¶

Display key statistics and quality metrics.

In [ ]:

# Display statistics
if "statistics" in results and results["statistics"]:
    print("Quality Statistics:")
    for metric, stats in results["statistics"].items():
        print(f"  {metric}: mean={stats['mean']:.3f}, std={stats['std']:.3f}")

# Show top instances
if "masks" in results and len(results["masks"]) > 0:
    print("\nTop 5 detected instances:")
    for i, mask in enumerate(results["masks"][:5]):
        print(
            f"  {i+1}. Instance {mask['mask_id']}: "
            f"IoU={mask['iou_pred']:.3f}, "
            f"Stability={mask['stability_score']:.3f}, "
            f"Area={mask['area']} pixels"
        )

# Display statistics if "statistics" in results and results["statistics"]: print("Quality Statistics:") for metric, stats in results["statistics"].items(): print(f" {metric}: mean={stats['mean']:.3f}, std={stats['std']:.3f}") # Show top instances if "masks" in results and len(results["masks"]) > 0: print("\nTop 5 detected instances:") for i, mask in enumerate(results["masks"][:5]): print( f" {i+1}. Instance {mask['mask_id']}: " f"IoU={mask['iou_pred']:.3f}, " f"Stability={mask['stability_score']:.3f}, " f"Area={mask['area']} pixels" )

Visualizations¶

Use the integrated visualization methods for comprehensive analysis.

In [ ]:

# probability visualization
detector.visualize_results(
    naip_2019_path,
    naip_2022_path,
    f"{out_folder}/binary_mask.tif",
    f"{out_folder}/probability_mask.tif",
)

# probability visualization detector.visualize_results( naip_2019_path, naip_2022_path, f"{out_folder}/binary_mask.tif", f"{out_folder}/probability_mask.tif", )

In [ ]:

# Create split comparison visualization
detector.create_split_comparison(
    naip_2019_path,
    naip_2022_path,
    f"{out_folder}/binary_mask.tif",
    f"{out_folder}/probability_mask.tif",
    f"{out_folder}/split_comparison.png",
)

# Create split comparison visualization detector.create_split_comparison( naip_2019_path, naip_2022_path, f"{out_folder}/binary_mask.tif", f"{out_folder}/probability_mask.tif", f"{out_folder}/split_comparison.png", )

In [ ]:

# Analyze individual instances
instance_stats = detector.analyze_instances(
    f"{out_folder}/instance_masks.tif",
    f"{out_folder}/instance_masks_scores.tif",
    f"{out_folder}/instance_analysis.png",
)

# Analyze individual instances instance_stats = detector.analyze_instances( f"{out_folder}/instance_masks.tif", f"{out_folder}/instance_masks_scores.tif", f"{out_folder}/instance_analysis.png", )

Comprehensive Analysis Report¶

Generate a detailed analysis report combining all metrics.

In [ ]:

# Create comprehensive analysis report
detector.create_comprehensive_report(results, f"{out_folder}/comprehensive_report.png")

# Create comprehensive analysis report detector.create_comprehensive_report(results, f"{out_folder}/comprehensive_report.png")

One-Click Complete Analysis¶

For ultimate simplicity, use the complete analysis method.

In [ ]:

# Alternative: Run complete analysis in one step
# This method does everything - detection, file outputs, and all visualizations
complete_results = detector.run_complete_analysis(
    naip_2019_path, naip_2022_path, "complete_analysis_results"
)

# Alternative: Run complete analysis in one step # This method does everything - detection, file outputs, and all visualizations complete_results = detector.run_complete_analysis( naip_2019_path, naip_2022_path, "complete_analysis_results" )

Summary¶

This notebook demonstrated the change detection functionality in GeoAI with integrated visualization methods:

Key Features Used:¶

1. Change Detection: Instance segmentation with confidence scoring
2. Integrated Visualizations: Built-in methods for comprehensive analysis
3. Simplified API: Clean, streamlined interface following geoai patterns
4. Complete Analysis: One-click method for full analysis workflow

Output Files Generated:¶

• binary_mask.tif: Traditional binary change detection
• probability_mask.tif: Probability-weighted change detection
• instance_masks.tif: Instance segmentation with unique IDs
• instance_masks_scores.tif: Confidence scores for each instance
• enhanced_probability_results.png: Comprehensive visualization
• split_comparison.png: Before/after split comparison
• instance_analysis.png: Individual instance analysis
• comprehensive_report.png: Complete analysis report

Advantages Over Traditional Methods:¶

1. Instance-Level Analysis: Each change object has unique ID and metrics
2. Quality Assessment: Confidence scores for filtering and ranking
3. Rich Visualizations: Multiple analysis perspectives in one package
4. Simplified Workflow: Integrated methods reduce code complexity
5. Flexible Usage: From simple detection to comprehensive analysis