Extreme Learning Machine with Composite Kernels for Hyperspectral Image Classification

Project Summary:

In this paper, based on ELM, we perform the joint spatial–spectral classification of HSIs. The fast computation capability of ELM can reduce the time for processing the high dimensionality and high-resolution HSI data. The good generalization performance of ELM helps to increase both the spectral and spatial separability and then leads to a superior performance by joint spatial–spectral classification. We propose two ELM-based CK spatial–spectral classification methods, namely, ELM with CK (ELM-CK) and kernel-based ELM with CK (KELM-CK). In ELM-CK, the single spatial or spectral kernel is represented as the multiplication of a random hidden layer output matrix and

its transposition. While in KELM-CK, the general Gaussian radial basis function (RBF) kernel is used. The composite spatial–spectral kernel is input to the ELM learning framework, resulting in a simple and effective solution for HIS classification.

Experimental results:



Yicong Zhou, Jiangtao Peng, and C. L. Philip Chen, “Extreme Learning Machine with Composite Kernels for Hyperspectral Image Classification,” IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing (JSTARS), vol. 8, no. 6, pp. 2351–2360, 2015.