Signal processing techniques for concealed weapon detection by use of neural networks

The use of active millimeter wave radar has proven successful in the field of Concealed Weapon Detection. Time resolved signals acquired from the radar scans are pre-processed and classified using an Artificial Neural Network. A problem with this method occurs in the training of the ANN, where the network must be trained on each weapon type that it is to reliably classify. Any deviation from the training weapon sub-class type leads to a decrease in the classifier´s performance. This is illustrated in the experimental results. A possible improvement on this method is outlined in the form of including signals reconstructed using Principal Component Analysis into the training set.