Photo Interpretation and Remote Sensing Module 10

This module's focus was the last method of image classification, supervised classification, which involves identifying known examples of different LULC classes so that the software can associate a spectral signature with each one and classify the remaining pixels based on their statistical similarity to the training classes. It's more work upfront than unsupervised classification, but there's the advantage of not having to go back and figure out what the classes are after the fact.

This was another fun lab assignment, but also kind of frustrating. Despite several attempts, I had a very hard time establishing a unique spectral signature for the road class, and as you might be able to see in my map below, even the final output has some confusion between the road and urban classes, and possibly also road and grass in some places.

You can also see from the distance output image (in which brighter areas correspond to areas that are more different from the established spectral signatures and thus more likely to end up misclassified) that there might be some issues with some of the agricultural areas and along the edges of certain features like the lakes. If this map were for a real-world application, those areas might need to be ground truthed or compared with other imagery to see if they need to be reclassified.

Photo Interpretation and Remote Sensing Module 9

This week we're back to land use land cover classification from aerial imagery, this time using automated methods. In this lab, we learned how to perform unsupervised classification in both ArcMap and ERDAS Imagine, culminating in a classified image of the UWF campus. For this assignment, the image was automatically classified into 50 different classes based on pixel characteristics, then manually simplified into five classes by identifying the broad LULC category each of the original 50 classes. My output is below; I ended up with a significant portion of the image falling into the "Mixed" class, which consists of pixels whose automatically assigned spectral class spanned multiple LULC classes (most of the confusion seemed to be between grass and buildings/roads--when I started out by assigning classes to the grass category, I suddenly had a number of green roofs!).