Classifications and Project Planning

Today I stayed back at the barn to work on classifying the WV2 imagery to identify and classify the major, pure stands of kuku’i all within the Ka’a’awa Valley. I learned a ton about eCognition today, primarily within segmentation, statistical extraction, feature extraction, and all sorts of other things within this fun program. Briton helped me out a ton all throughout the morning working constantly towards improving the product that I was creating. I corrected the past 2 days GPS points with the local Oahu base station for maximum accuracy and found that I achieved much better precision with the older trimble units which was surprising. However, all of my accuracy was sufficient for vegetation stand identification.

I mapped all of this in ArcGIS, and cleaned up the attribute tables and renamed pictures for attribute association. These ground truthing points are essential in supervised object based statistical classifications in eCog. The majority of the rest of my day was spent working on eCognition in multiple different classifications and segmentations. This program offers limitless authority to the user to determine their settings which reflect in the final product. I kept record of differing scale parameters, shape and compactness threshold controls, layer weights, and more all working towards building the best segmentation which would come into play later. Because the focus of my classification is kuku’i, the identification of this species is the primary goal which I worked towards. Statistical classification did not give sufficient power for analysis, and Kerry and Jessica helped show me the ropes of the feature extraction classification method which is in my opinion exponentially more effective than a general statistical classification with sample site selections. I was very glad to learn this method and will continue improving my classification tomorrow. Feature extraction is conducted by user control and examination of values per band for the various land cover types, see the attached picture where I planned out and calculated differences between classes to determine band settings for extraction and classification.

While I had been working all day on classifications, I was not one hundred percent on my final topic, and replanned the second half of my project with Dr. Wechsler. My new plan is to compare my kuku’i classification from WV2 in eCognition to digitized polygons of kuku’i regions from UAV imagery that has been mosaic-ed into orthophotos. Within ArcGIS, areas of overlap, intersect, can be compared against total areas of union, and areas can be quantified. From all this, the percent coefficient of areal correlation can be calculated to evaluate classification strategies. Ideally I will be able to also classify UAV mosaics but I will see how far I can get with the current plan tomorrow and next week. Overall a productive day and I will really be on a roll the next day or two when I can improve my WV2 classification and obtain my mosaics of the 2 study sites. Another great day with the CSULB GRAM program in Hawaii!!