IMAGE CLASSIFICATION
Problem
The Black Water National Wildlife Refuge is looking to classify land into different categories in order to develop a management plan. The refuge managers have provided multispectral aerial photography to be used in conducting landuse classification.
​
Analysis Procedures
The data available for this procedure is a 4-band aerial photograph collected in August of 2010 that has a spatial resolution of 1 foot. The 4 bands correspond to wavelengths in the red, green, blue, and near-infrared (NIR) spectra. To classify the image for different land cover categories, I plan to use a false-color composite consisting of the red, green, and NIR bands and conduct a supervised classification.
​
For this analysis, I used ArcGIS Pro. I began by opening the aerial photography and displaying it as a false-color infrared. Next, I created roughly three polygons for each of the six following classes: water, developed, barren, forest, cultivated, or wetlands. I then saved the polygons as a shapefile and used them to run a supervised classification.
​
The resulting raster was exported to a TIFF file, and I then used the Build Raster Attribute Table tool to create an attribute table that included cell counts for each class. I then created a new field and calculated the total area for each class by multiplying the cell count by 0.092903. Because the cells were each 1 square foot, this value was used to convert from square feet to square meters.
​
After completing this supervised classification, it was apparent that the results were not accurate, so additional polygons were created for each class, and a secondary supervised classification was conducted.
Figure 1. Diagram of methods.
Results
.jpg)
.jpg)
.jpg)
.jpg)
Application & Reflection
Image classification from an agricultural perspective is one of the primary reasons I became interested in GIS. It can be used to differentiate between crop and weed species, or between healthy and unhealthy plants.
​
Problem Description: As an example, image classification can be used to classify an agricultural field into areas of healthy crop stand and areas of nutrient deficiency.
​
Data Needed: To conduct this type of analysis, one would likely desire to use multispectral aerial photography. Some of this data may be found on websites such as USGS EarthExplorer, but it would also be good to obtain one’s own data. Many UASs with multispectral sensors can readily be purchased or rented, and can then be used to fly over an agricultural field. In addition to the aerial photography, it would be important to obtain ground truth data, such as foliar nutrient analysis of healthy and nutrient deficient plants. That way, you can have some reference points to use for your classification.
​
Analysis Procedures: To conduct this classification, one would have to create training samples representing bare earth, healthy plants, nutrient deficient plants, and potentially weeds or other subjects depending on what is in the field of view. Then, you could use the training samples to run a supervised classification. Lastly, one could compare the classified raster with the aerial photography and the ground truth data to ensure that the classification was accurate.