Object-oriented approach of information extraction from panchromatic satellite images based on fuzzy logic

This paper presents a fuzzy inference based object oriented classification approach for information extraction from panchromatic satellite images. The main problem concerned with high resolution satellite imagery is diverse composition of spectral resolution in urban mapping. Information extraction form these high resolution images are problematic due to discriminated object boundaries in the land use as well as the multiple resemblances of the different class segments. Satisfying results cannot be obtained with the use of traditional pixel based classifiers which rely on per pixel value analysis. This paper demonstrated the concept of fuzzy logic based object-oriented approach, in order to deal with this problematic situation. The concept of this approach is to determine a suitable threshold of the parameters adopted in the object oriented approach. Establishment of fuzzy membership functions for extraction of various surface objects such as bare soil, roads, water, vegetation, building etc using the provided built-in tools of eCognition software, is the main focus of this study. In this proposed approach, fuzzy logic (rules) is defined for every class based on the specifications mentioned in class hierarchy. Using these fuzzy rules each image object is tested against specifications. Each segment receives a membership value (degree) which acts as class representative. After deffuzification of the results every class is assigned associated segments with high similarity degree, using eCognition software. Improved overall accuracy of the presented system, making this approach feasible and efficient for extracting information from high resolution remotely sensed data.