关键词: 极限学习机, 高光谱遥感, 分类, 参数优化

Abstract:

Machine learning technology has been widely used in remote sensing image classification. Extreme learning machine (ELM) is proposed recently for image classification, but the regularization and kernel parameters (C, σ) of ELM have significant influence on classification performance. In this paper, an ELM classifier with firefly algorithm (FA)-based parameter optimization is proposed for hyperspectral image classification. Firstly, FA algorithm is used for band selection in order to reduce the computational load of hyperspectral image classification. Then, the parameters (C, σ) of ELM are optimized by FA with respect to the classification accuracy. In our experiments, the firefly algorithm is also compared with other parameter optimization algorithms such as genetic algorithm (GA) and particle swarm optimization (PSO). In addition, the support vector machine (SVM)-based classification algorithm is also implemented for comparison purpose. The experiments are conducted on three classical hyperspectral remote sensing data. Results indicate that the performance of ELM method is better than SVM method from the aspects of classification accuracy and running time. Our experiments successfully prove that the proposed algorithm can provide a better performance for hyperspectral image classification.

Key words: ELM, hyperspectral remote sensing, classification, parameter optimization