In Exercise 1 of the lab, ArcGIS Pro was utilized for unsupervised classification of a high-resolution aerial photograph of the UWF campus. The procedure started with a new project and a map, incorporating the aerial image featuring three visible bands: Red, Green, and Blue (RGB), along with a a lower spatial resolution Landsat 9 image to include Red and Infrared bands. Through the Classification Wizard, an unsupervised classification was executed, setting a maximum of 30 classes and 20 iterations, yielding a temporary layer with cell values classified from 1 to 29. These classes were consolidated into five classifications: Trees, Grass, Buildings/Roads, Shadows, and Mixed, the last of which was designated for overlapping categorizations.
Further refinement was achieved using Layer Blend and Transparency techniques with the infrared image. A new layer reflecting these five classifications was added, and an “Area” field was introduced in the attribute table to compute the area for each class. The area was calculated using the formula (Count * (1*1)), converting to acres with (!Count! * 0.00024711).
In exercise 2, a training sample for Supervised Classification is created using data from Landsat 9 (OLI-2). Three band combinations are set: False Natural Color (bands 6, 5, 4), True Color (bands 4, 3, 2), and False Color IR (bands 5, 4, 3). The process begins in the Training Samples Manager under the Imagery tab in Classification Tools, where two new classes—Clear cut and Estuary—are added, resulting in six classes: Water, Developed, Forest, Planted, Estuary, and Clear cut. Approximately ten polygons are delineated for each class using polygons or freehand tools.
After class delineation, the evaluation phase starts by creating charts accessed from the Data tab through the Spectral Profile feature. These charts assist in assessing overlaps among training samples of different classes. Overlapping samples are tracked using their OBJECTID identifiers corresponding to the polygons. Training samples are exported, excluding those excessively overlapping with other classes' samples.
The final step incorporates machine learning techniques to generate classified images using the training samples. The classification process is executed via the Classify option in the Imagery tab, employing four classification methods: Maximum Likelihood, Random Trees, Support Vector Machine, and K-Nearest Neighbor. The four classification types are evaluated to select the preferred method, calculate the area for each class, and layout map is created.







