Determining Land Cover using Object Based Classification

Introduction

Fig 1: Red Edge Sensor

  The reclassification done in this report will use data collected from a Red Edge sensor shown in Figure 1.A red edge sensor has 5 bands. The band information can be seen below in Figure 2. In order of wavelength, the correct order when creating a composite image is blue, green, red, red edge, and Near IR (NIR).

Fig 2: Red Edge Spectral Bands

  The area looked at is near south middle school ponds are located in Eau Claire Wisconsin. ArcGIS's spatial analyst extension is heavily used in the reclassification of this area. There are five main steps when performing the reclassification:

• Create a composite image using the Composite Bands tool
• Extract bands by using the Segment Mean Shift tool
• Create training samples using the training sample manager from the image classification toolbar
• Create a definition file using the Train Random Trees Classifier or Train Support Vector Machine Classifier tool
• Determine surface types  using the Classify Raster tool
• Use the Reclassify Raster tool to group surface types together into land cover subtypes
• Create assessment points with the Create Accuracy Assessment Points tool
• Use the editor toolbar and original RGB imagery to determine ground truth values
• Create a confusion matrix using the Compute Confusion Matrix tool to see how accurate the reclassification.