TEMPORAL WINDOW SYSTEM - A NEW APPROACH FOR DYNAMIC DETECTION - APPLICATION TO SURFACE-ACOUSTIC-WAVE GAS SENSORS

The use of artificial neural networks-based pattern recognition techni ques is now frequent and efficient in the gas sensor signal processing domain. Neural network data sets are generally built with steady stat e sensor responses. Nevertheless, when the detection speed is an essen tial parameter, they must be monitored in a real time mode. In this pa per, we present a new dynamic approach and illustrate it with surface acoustic wave sensor NO2 responses. A filter and a detector constitute the system. They are both implemented with neural networks and both u se shifting temporal windows, The filter is based on a recirculation f irst neural network whereas a backpropagation second neural network is used for the detector whose output is compared to a threshold to turn on an alarm. The aim is to realize a smart sensor. Since no theoretic al results exist yet to find the optimal size, initialization and para metrization of backpropagation neural networks, we have studied the in fluences of weight initialization, temporal window width, hidden neuro n number and learning rule parameters. The results show that the best convergence speeds are obtained for weight initial values that are ver y dependent on the network topology. It was also found that the learni ng quality and generalization properties were independent of the weigh t initialization. Other results on the network architecture and perfor mances are presented and discussed.