Authors: LASYA VENIGALLA*, University of Texas, Dallas, Fang Qiu, Professor and Head of the Department, Geospatial Information Sciences, University of Texas, Dallas
Topics: Remote Sensing
Keywords: Object-based Classification, Neuro-Fuzzy classifier, Within-object spectral variability
Session Type: Paper
Presentation File: No File Uploaded
The increased capabilities of a remote sensing sensor resulted in a wide range of a high-resolution image (HRI). The increase in resolution makes it possible to view the minute details on the earth's surface. However, while identifying the features pixel-by-pixel, only spatial properties of the image are considered and textural properties are not considered. Therefore, object-based classification techniques that take both spatial as well as textural properties into consideration are used to extract features from a high-resolution image. However, rule-based methods that are used for object-based classification consider single summary statistic value to represent the entire object and don’t consider the within object spectral variability. There are curve-based matching methods that utilize all the pixels inside an object, but matching is done on each individual sample, lacking generalizing ability. In this paper, both within-object spectral variability and generalizing ability are considered by treating histograms of the objects as spectral curves and performing neuro-fuzzy classification using Gaussian-Fuzzy Learning Vector Quantization (GFLVQ) algorithm. The study is also compared with Kolmogorov–Smirnov (KS) distance function which is a curve-based matching algorithm and resulted in improved accuracy of 92%.