Search

Securing global trade requires efficient screening of containers for threat materials. This work demonstrates a novel approach combining fast neutron activation analysis and artificial neural network (ANN) to identify elemental composition of sealed cargo, in particular elements carbon, oxygen, and nitrogen, which are the main components of explosives. This study shows that Rapidly Relocatable Tagged Neutron Inspection System in combination with ANN is a potential promising solution for the inspection of sealed containers, allowing precise identification of elements and detection of potential threats without the need to open the containers. [ABSTRACT FROM AUTHOR]

In this paper, we propose an approach to determine the time constants and the amplitudes of the mass loading effect and of the viscoelastic contribution of SAW sensor’s frequency shift. This approach consists in optimizing a function of these parameters, which is independent of the concentration profile. We experimentally establish in laboratory conditions ( $T$ = 22 °C), on a data set composed of seven different gases, that these features are suitable for chemical compounds identification. In particular, we obtain a higher classification rate than the traditional amplitudes of the signals during the steady state, and we show that the classification success rate can be increased by using both of them in conjunction with a feature subset selection heuristic. We also propose a method based on deconvolution and kernel regression to estimate the temporal concentration profile. [ABSTRACT FROM PUBLISHER]

Holders of extensive television archives have to describe television streams which are composed of various telecasts. The first step to this description is to break the stream into small logical units like telecasts or advertisements in order to describe each of them separately. In this article we focus on the television stream structuring which consists in finding automatically the beginning and end of the various telecasts. Since program guides do not include the real content of television streams, we propose an approach based on the modeling of television schedules. We aim for predicting all the possible television schedules for a particular day, so that the automatic system then needs only a few detections to obtain the accurate boundaries of each part of the stream. [ABSTRACT FROM AUTHOR]