The increased availability of high-resolution satellite imagery presents many new opportunities for detailed classification of urban land use and land cover, but it also brings to light the inadequacies of traditional spectral-based classification methods using the maximum-likelihood classifier. Recent research in remote sensing tends to focus on developing new textural methods and alternative classifiers to address this problem, but the potentially promising possibility of optimizing the maximum-likelihood classifier and combining multiple textural characteristics for image classification remains unexplored. Moreover, as an important parameter in texture analysis, the appropriate size of moving window (critical window) used to extract different textural layers and its association with land-use and/or land-cover objects has generated a lot of research, but the quantitative principles remain unclear. The overall goal of this doctoral dissertation research project is to develop an efficient, simple, and robust approach for image classification that has potential for automation. The doctoral candidate's objectives are (1) to develop and test a genetic Bayesian classifier to efficiently boost the performance of the traditional maximum-likelihood classifier by optimizing the prior probabilities with a genetic algorithm; (2) to determine whether a combination of textural indices, such as fractal dimension, lacunarity, and Moran's I, is efficient in classifying heterogeneous urban area, using the genetic Bayesian classifier; and (3) to derive the approximate size of the "critical window" when dealing with these texture indices and high-resolution IKONOS imagery. This project aims to improve urban land-use/land-cover classification accuracy by applying the new approach to high-resolution IKONOS pre- and post-Katrina imagery in New Orleans.
The results of this study will provide a new method to improve land-use/land-cover classification through addressing cutting edge issues like genetic algorithm, Bayesian approach, critical window, fractal geometry, and spatial autocorrelation. The analysis of critical window and texture vector will provide better understanding of the relationships between the texture measures, land-use objects, and sensor resolutions, all of which are critical to the success of texture-aided classification. The emphasis on the development of a simple, quick, and reliable classification method is imperative for future automation and many applications such as rapid monitoring of disastrous and special events. The study area, New Orleans, was severely inundated by Hurricane Katrina in 2005. This project aims to provide a better post-disaster assessment with high-resolution satellite imagery that could help in guiding the recovery and urban planning. As a Doctoral Dissertation Research Improvement award, this award also will provide support to enable a promising student to establish a strong independent research career.
