Investigating the Use of Pixel-Based vs Object-Based Approaches for Supervised Aerial Image Classification
Michael Gover
Co-Presenters: Individual Presentation
College: The Dorothy and George Hennings College of Science, Mathematics and Technology
Major: Earth Science Teacher Certification Option
Faculty Research Mentor: Feng Qi
Abstract:
Characteristics of the environment affects our psychological and general well-being. This research is a part of a larger project comparing the environmental properties derived from street view images using image segmentation and those derived from aerial images using image classification. In our efforts to derive environmental properties from aerial images we explore the strengths and challenges with using Supervised Pixel-based image classification and Supervised Object-based image classification in ArcGIS Pro. We analyzed the environmental properties of two aerial images of North New Jersey neighborhoods where our Google Street View images are located. We created training samples and ran several different classifier programs that ArcGIS offers for both Pixel-based and Object-based image classification and compared the results. Pixel-based was limited in its ability to differentiate between ambiguous pixels and resulted in specky surfaces requiring extensive amounts of manual reclassification. Object-based required extra steps to create effective training samples but created much cleaner classified images that only required some reclassification. Object-based classified images were used to derive environmental properties within fix-distance buffers around random locations in the two study sites with Zonal Histogram. Through continued research we hope to continue analyzing the environmental properties we derived from our classified aerial images and compare our results with those derived from our street view image segmentation.