Tutorial Geobia for ArcGIS

Summary

The tutorial demonstrates object-based image analysis in ArcGIS using two toolboxes:

• Landsat 8 toolbox: Preprocesses Landsat 8 imagery and includes image-segmentation tools.
• GeoBIA toolbox: Supports object-based analysis of polygon layers and is not limited to Landsat 8 imagery.

Workflow

1. Prepare the data. Download and unzip a Landsat 8 scene. The example uses imagery from Central Europe, acquired on September 18, 2014 (path 189, row 23).
2. Define an area of interest (AOI). Create a shapefile containing a single rectangular feature. The tutorial begins with an AOI of about 15 × 15 km, noting that its size is limited by available computer memory. The shapefile’s coordinate reference system should match the imagery.
3. Create a composite. Run the composite tool with a workspace, Landsat metadata file, AOI, project name, and output resolution. The example uses a resolution of 50 m and does not apply a smoothing filter. Display the resulting composite with different band combinations and a standard-deviation stretch to view natural color.
4. Apply radiometric and atmospheric processing. The tools generate image products and a text file listing image filenames and rescaling information. Digital-number values are converted to radiance using Landsat metadata coefficients, then atmospheric correction is performed using dark object subtraction, assuming 1% minimum reflectance.
5. Segment the image. Use the segmentation tool to create a polygon shapefile of image objects. Select a raster edge threshold and inspect the result. A threshold of 30 misses some important boundaries, so the tutorial tests 15, which produces a more satisfactory segmentation. The final run takes about 33 minutes.

Practical takeaway

Inspect the segmentation output and adjust the edge-threshold setting before running the full process. The tutorial concludes by displaying the resulting object polygons.

Speaker and source

A single narrator guides viewers through the ArcGIS workflow; no other speakers or sources are identifiable.