Classification of Air-Written Letters Using RF Signals.

In this study, a method for non-contact sensing and classification of 26 air-written letters using radio-frequency (RF) signals is proposed. Software Defined Radio (SDR) modules are used for RF waveform transmitting and receiving. Two perpendicular dipole antennas are used in the receiver for the polarization diversity which provides better accuracy. The proposed approach uses Discrete Cosine Transform (DCT) coefficients for the distinctive features and the feature size is also reduced by applying Principal Component Analysis (PCA). The Support Vector Machines (SVM) classification algorithm is chosen for best accuracy. It is shown in the experiments conducted with real measurements that the proposed method improves the classification performance of 26 air-written letters with an overall accuracy of 93.77% for 20 features.