UAS-based Change Detection

Center for Geospatial Analytics at North Carolina State University

Corey White

Outline

  • Selection of spatial-temporal resolution
  • Determination of processing time step
  • Sensor and environmental considerations
  • Selection of classification schema
  • Feature engineering
  • Data Sampling
  • Selection of classification and change detection methodology

What is Change Detection?

Why does it Matter?

  • White CT, Reckling W, Petrasova A, Meentemeyer RK, Mitasova H. Rapid-DEM: Rapid Topographic Updates through Satellite Change Detection and UAS Data Fusion. Remote Sensing. 2022; 14(7):1718. https://doi.org/10.3390/rs14071718

Process of Interest

  • What process are you interested in montioring?
    • Is it naturally occuring?
    • Is it man made?
    • Is it fast changing?
    • Is it slow changing?

Spatial Resolution - Continous Data

Continous Data

  • Corey White, & Petras, V. (2024). tomorrownow/intro-to-geoprocessing-workshop: Zendo Release (Version 0.1.1) [Computer software]. https://doi.org/10.5281/zenodo.10764926

Spatial Resolution - Discrete Data

Discrete

  • Corey White, & Petras, V. (2024). tomorrownow/intro-to-geoprocessing-workshop: Zendo Release (Version 0.1.1) [Computer software]. https://doi.org/10.5281/zenodo.10764926

Temporal Resolution

Minutes

Continous: Depth (m)

Years

Discrete: Land Cover Change

Process Time Step

Phenological considerations

  • Natural variation (e.g. seasonal, extreme weather event)
  • Anthroprogentic change (e.g. construction, harvest)

Discrete Time Step

  • Hourly
  • Daily
  • Weekly
  • Monthly
  • Seasonally
  • Annually
  • Bi-annually

Continous

  • Live video feed

Sensor and Environmental Considerations

Sensor

  • Spatial Resolution
  • Spectral Resolution
  • Radiometric

Environmental

  • Sun azimuth
  • Sun elevation
  • Cloud coverage
  • Off-nadir
  • Temporal resolution
  • Phenological cycles

Tip

Best if conditions are as similar as possible

Classification Schemas

  • Land Cover/Land Use
    • NLCD (National Land Cover Dataset)
    • NWALT (U.S. Conterminous Wall-to-Wall Anthropogenic Land Use Trends)
  • Custom

https://www.usgs.gov/centers/eros

Spectral Bands

RGB

red green blue

Mulispectral (N-separated bands)

Hyperspectral (Continous Spectrum)

Feature Engineering

  • Normalized Indices
    • NDVI, NDWI, VARI, etc..
  • Low and High Pass Filters
    • Low Pass - Smoothing Kernal
    • High Pass - Edge Detection
  • Textural features
    • Grey level co-occurrence matrix (GLCM)
      • Angular Second Moment
      • Contrast
      • Correlation
  • Principal component analysis (PCA)

VARI \[ VARI = \frac{(Green - Red)}{(Green + Red - Blue)} \]

Sampling Methods

  • Random Sample
    • Sample the entire spatial extent \(n\) number of locations
  • Cluster Sampling
    • Sample by different clusters
  • Stratified Random Sample
    • Sample by different classes (e.g. Developed, Undeveloped)

Classification and Change Detection

  1. Change Detection Method
    • Binary / Thematic
  2. Image Classification
    • Supervised
    • Unsupervised
  3. Classification Validation
    • Confusion Matrix
    • Overall Accuarcy
    • Kappa
  4. Thematic Change

Binary Change Detection

Identify Change or no change

Methods

  • Visual change detection
  • Thresholds (e.g. Image Differencing)
  • PCA (Principle Component Analysis)
  • Logistic Classification

Thematic Change Dection

  • Pixel-based Post-Classification Thematic Change Detection
  • Object-based Post-Classification Thematic Change Detection

Pixel-based

  • Medium to low resolution data
  • Track per-pixel changes over time

Object-based

  • Very high resolution (VHR) data (UAS data!)
  • Image Segmentation
  • Track objects over time

Classification Methods

Supervised Classification

  • Needs labeled training data

Unsupervised Classification

  • No training data needed
  • Classify the results

Classification Methods

Supervised Classification Algorithms

  • Random Forest
  • Support Vector Machine (SVN)
  • Maximum Likelihood Classification (MLC)
  • Artifical Neural Networks (ANN)

Unsupervised Classification Algorithms

  • K-Means Clustering
  • Iterative Self-Organizing Data Analysis (ISODATA)
  • Spectral Change Vector Analysis

Classification Results

Model Validation

Confusion Matrix

Reference  Predicted Forest Water Urban Agriculture User’s Accuracy (%)
Forest 50 2 3 0 90.9
Water 1 45 0 4 88.2
Urban 2 1 40 7 80.0
Agriculture 0 3 5 42 85.7
Producer’s Accuracy (%) 94.3 86.5 83.3 79.2
  • \(Overall\ Accuarcy = \frac{Correctly\ classified\ samples}{Total\ samples}\)
  • Kappa coefficient (Cohen’s Kappa): inter-rater agreement

Post-Classification Change Detection

flowchart TB
  A[Image - Date 1]
  B[Image - Date 2]
  A --> AF[Feature Engineering]
  B --> BF[Feature Engineering]
  AF -- Pixel or Object --> AC[Classification]
  BF -- Pixel or Object --> BC[Classification]
  AC --> CM[Change Map]
  BC --> CM

Thematic Change (from-to)

  • White CT, Reckling W, Petrasova A, Meentemeyer RK, Mitasova H. Rapid-DEM: Rapid Topographic Updates through Satellite Change Detection and UAS Data Fusion. Remote Sensing. 2022; 14(7):1718. https://doi.org/10.3390/rs14071718