Week 10: Supervised Classification

Week 10 objectives are to create spectral signatures and AOI features in ERDAS.  From there, create a newly classified image from a satellite image.  Once the image is recoded, identify and resolve confusion between the signatures.

Create the map below, I started with an image of Germantown, Maryland.  I opened the signature editor and added a new AOI layer.  I used the polygon tool to select features such as water and roads.  I used the inquire tool and coordinates to find urban areas, agriculture, and more.  If the feature seemed to contain a lot of differing pixels, I used the growing properties to select the feature I used the inquire tool for and adjusted the spectral Euclidean Distance. 

Once all my signatures were created, I looked at the different layers to see where confusion existed and to identify the best bands to display.  I used bands R-4, G-5, and B-6.  The final step was to recode all of the signatures.  I was able to calculate the areas of each class within ERDAS.  I was not 100% satisfied I was not able to eliminate all pixel confusion.  I attempted to adjust my signatures and change bands, but did not seem to find success.  I look forward to applying these techniques in through a potential job or practice to improve my skills.

Supervised Classification of Germantown, Maryland using bands R-4, G-5, and B-6

Module 9: Unsupervised Classification

Unsupervised Classification

The goal of the lab was to perform an unsupervised classification using ArcMap and ERDAS.  Another aspect of the lab was to classify images with different spatial and spectral resolution.  The final goal was to learn how to reclassify and recode images in ERDAS.

The lab started off with using the Iso Cluster tool and Maximum Likelihood Classification tool.  Doing this, it created a classified image which I assigned classes colors.

In ERDAS, I used the Unsupervised Classification tool.  This allowed for me to give the image 50 classes.  Once the image had the classes, I reclassified by opening the attribute tables to change the colors of the pixels that belong to each feature.  I did this throughout using the Swipe tool to help identify from the true image.  It was also helpful to change the pixel group to a red or yellow to see how much it is used in the image.  From there, I continued to classify the pixels.

Once the pixels were classified, I used the Merge tool to group the 5 classifications into 5 classes.  I looked at the classes and assigned them groups to bring the classes from 50 to 5. I added the area of the features to help determine how much is impermeable and permeable classes.