Urban land use and land cover classification using the neural-fuzzy inference approach with Formosat-2 data

Abstract

This paper presents a neural-fuzzy inference approach to identify the land use and land cover (LULC) patterns in large urban areas with the 8-meter resolution of multi-spectral images collected by Formosat-2 satellite. Texture and feature analyses support the retrieval of fuzzy rules in the context of data mining to discern the embedded LULC patterns via a neural-fuzzy inference approach. The case study for Taichung City in central Taiwan shows the application potential based on five LULC classes. With the aid of integrated fuzzy rules and a neural network model, the optimal weights associated with these achievable rules can be determined with phenomenological and theoretical implications. Through appropriate model training and validation stages with respect to a groundtruth data set, research findings clearly indicate that the proposed remote sensing technique can structure an improved screening and sequencing procedure when selecting rules for LULC classification. There is no limitation of using broad spectral bands for category separation by this method, such as the ability to reliably separate only a few (4-5) classes. This normalized difference vegetation index (NDVI)-based data mining technique has shown potential for LULC pattern recognition in different regions, and is not restricted to this sensor, location or date.