Hyperspectral imagery has been widely used in terrain classification for its high resolution. Urban vegetation, known as an essential part of the urban ecosystem, can be difficult to discern due to high similarity of spectral signatures among some land-cover classes. In this paper, we investigate a hybrid approach of the genetic-algorithm tuned fuzzy support vector machine (GA-FSVM) technique and apply it to urban vegetation classification from aerial hyperspectral urban imagery. The approach adopts the genetic algorithm to optimize parameters of support vector machine, and employs the K-nearest neighbor algorithm to calculate the membership function for each fuzzy parameter, aiming to reduce the effects of the isolated and noisy samples. Test data come from push-broom hyperspectral imager (PHI) hyperspectral remote sensing image which partially covers a corner of the Shanghai World Exposition Park, while PHI is a hyper-spectral sensor developed by Shanghai Institute of Technical Physics. Experimental results show the GA-FSVM model generates overall accuracy of 71.2%, outperforming the maximum likelihood classifier with 49.4% accuracy and the artificial neural network method with 60.8% accuracy. It indicates GA-FSVM is a promising model for vegetation classification from hyperspectral urban data, and has good advantage in the application of classification involving abundant mixed pixels and small samples problem.
Hyperspectral remote sensing data has been widely used in Terrain Classification for its high resolution. The
classification of urban vegetation, identified as an indispensable and essential part of urban development system, is now
facing a major challenge as different complex land-cover classes having similar spectral signatures. For a better accuracy
in classification of urban vegetation, a classifier model was designed in this paper based on genetic algorithm (GA) and
support vector machine (SVM) to address the multiclass problem, and tests were made with the classification of PHI
hyperspectral remote sensing images acquired in 2003 which partially covers a corner of the Shanghai World Exposition
Park, while PHI is a hyper-spectral sensor developed by Shanghai Institute of Technical Physics.
SVM, based on statistical learning theory and structural risk minimization, is now widely used in classification in many
fields such as two-class classification, and also the multi-class classification later due to its superior performance. On the
other hand as parameters are very important factors affecting SVM's ability in classification, therefore, how to choose
the optimal parameters turned out to be one of the most urgent problems. In this paper, GA was used to acquire the
optimal parameters with following 3 steps. Firstly, useful training samples were selected according to the features of
hyperspectral images, to build the classifier model by applying radial basis function (RBF) kernel function and decision
Directed Acyclic Graph (DAG) strategy. Secondly, GA was introduced to optimize the parameters of SVM classification
model based on the gridsearch and Bayesian algorithm. Lastly, the proposed GA-SVM model was tested for results'
accuracy comparison with the maximum likelihood estimation and neural network model. Experimental results showed
that GA-SVM model performed better classified accuracy, indicating the coupling of GA and SVM model could
improve classification accuracy of hyperspectral remote sensing images, especially in vegetation classification.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.