Image Classification
Problem
The objective of this exercise is to demonstrate an understanding of image classification. Given an aerial photograph of Black Water Wildlife Refuge, the problem is to create a raster which displays the land cover classification using six categories: forest, cultivated field, barren area, developed/impervious, wetland, and water.
Analysis Procedures
I used the Esri module, “Visualizing and Analyzing Imagery with ArcGIS 10” to help me with understanding the procedures for this exercise. I used the Spatial Analyst toolbar to run the supervised classification of the orthophoto provided, specifically using a display of the infrared band to aid in classification.
The data used in this exercise is an orthophoto of the study area, Black Water Wildlife Refuge. The file contains three bands, including the infrared raster. First, I changed the appearance of the image from the true-color to color-infrared by changing how the bands were set up. I set band 4 as “Red”, band 3 as “Green”, and band 2 as “Blue”. Next I performed a supervised classification in two parts.
Part 1. I performed a supervised classification by using at least two sample polygons per Land Cover Class. More than two sample polygons were used in cases when there were many different colors within the same Land Cover Class category. I used small polygons in order to attain homogeneous samples. I saved the sample polygons as a shapefile and performed the interactive supervised classification on these samples. I found that while most of the area was correctly classified, there were many misclassified areas. These areas were often misclassified as either wetlands or developed/impervious area, while there were also cultivated fields partially classified as forest.
Part 2. I re-ran the analysis described in Part 1. This time I used more numerous and smaller polygons, in effort to keep each polygon uniform while collecting an area for each important color in the classification. I created a shapefile and performed the interactive supervised classification. The classification raster generated in step 2 had much more differentiation within areas which appear to be a continuous class by the orthophoto. For example, there was much more wetland area mixed into the southern forest region, and a shaded part of the barren region (evident in the orthophoto) which was classified partly as developed and partly as wetland in the original classification was classified much more heavily as this combination in the second classification attempt.
Results
The resulting maps are shown below. Polygons used to create each map are also shown.
The resulting maps are shown below. Polygons used to create each map are also shown.
Application and Reflection
This procedure seems to be a very interesting and applicable one. In my research on recreational trails, trails can be classified in different ways based on their condition (trampled, stunted vegetation, baren, etc). I am interested in testing the land use classification system on trails to determine if orthoimagery could be used to determine the condition classes of trails. This type of classification would be much finer scale than the area we classified in this exercise, but if imagery could be attained in high enough detail, the procedure may be applicable. Another way I can apply this to my work is to identify disturbed regions in recreational areas, such as meadows. Applying this procedure could save a lot of time and money in field work.