Supervised Feature Selection Method for High-Dimensional Data Classification in Photo-Thermal Infrared Imaging with Limited Training Data

We study the kernel based SVM algorithm with variable models to adapt to the high-dimensional but relatively small samples for remote explosive detection on photo-thermal infrared imaging spectroscopy (PT-IRIS) classification. The response plot, predicted vs. actual plot, and residuals plot of the linear, quadratic, cubic, and coarse Gaussian SVM are demonstrated. In addition, a comprehensive comparison of classification performance of these SVM models is conducted in terms of root mean square error, R-squared, mean squared error, and mean absolute error. The excellent experimental results demonstrated that the kernel based SVM models provide a very promising feature selection solution to high-dimensional data sets with limited training samples.