Radar classification of indoor targets using support vector machines

Through-the-wall radar systems are typically plagued by harsh clutter which confounds the detection and classification of targets. This study approaches the problem of separating humans from indoor targets through machine learning techniques by implementing support vector machine (SVM) classifiers. Simulated and experimental data are used to evaluate the suitability of SVMs for target classification via their spectral characteristics. Numerical modelling of humans and indoor clutter was accomplished through the finite-difference time-domain (FDTD) method. The radar cross-sections for the various objects are acquired over a wide range of frequencies, polarisations, aspect angles, and material properties. Furthermore, the spectral properties of humans and clutter targets are measured using a network analyser. The potentials of the acquired simulated and experimental spectral characteristics at different frequencies and polarisations are explored for target classification using SVMs. Feature selection algorithms are also investigated to reduce the redundancy and model complexity. Finally, the classification performance is assessed in the presence of additive white Gaussian noise and through various wall materials.