Created with Sketch.
Hyperspectral data collection for the assessment of target detection algorithms: the Viareggio 2013 trial
Hyperspectral sensors mounted on board airborne platforms collect spectral data that can be exploited for several applications, such as classification, target identification, detection of anomalies and changes occurred within multiple acquisitions. In the framework of target detection, the performance assessment of different algorithms in real cases is a very important and critical issue for both the user and the scientific communities. Experimental performance assessment needs the availability of extensive data sets equipped with welldocumented ground truth and designed to properly represent different operating scenarios. The lack of public available hyperspectral data collection for target detection algorithms assessment has motivated us in performing an extensive measurement campaign including various operating scenarios. It was organized by CISAM (Centro Interforze Studi e Applicazioni Militari) in cooperation with University of Pisa, Selex-ES and CSSN (Naval Experimentation and Support Centre), and it was conducted in the city of Viareggio, located in the north-west coastline of Tuscany, Italy on 8-9 May, 2013. The airborne Selex- ES hyperspectral sensor SIMGA was adopted for data collection. SIMGA sensor is equipped with two high spectral resolution cameras, operating in the Visible-Near Infrared (VNIR) and in the Short Wave Infrared (SWIR) spectral regions. Hyperspectral data were collected over different sites at morning and afternoon of two subsequent days. This paper describes the large multi-temporal hyperspectral data collection acquired during the trial. The data were collected to exhaustively assess the performance of anomalous change detection, anomaly detection and spectral matching techniques. Four different sites were set up, representing a complex urban scenario, two parking lots and a coastline/rural area. Part of a selected parking lot site was dedicated to calibration purposes. For the urban and parking lot areas, cars and opaque cloths used to cover cars were used. For the rural scenario, composed of beach and Mediterranean vegetation, polystyrene panels covered by different materials were deployed on the beach, while common clothes were arranged within the vegetation. Generally, the dimension of the deployed targets, in terms of the sensor ground resolution, ranged from a dozen pixels, for largest targets as cars or vans, to less than a pixel, for smallest targets, such as clothes. An extensive ground truth documentation was carried out to give evidence of targets of opportunity. Experiments to test anomalous change detection techniques were set up changing the geographical position of the deployed materials and vehicles in a urban scenario. Search and rescue operations were simulated to evaluate the performance of anomaly detection algorithms. For this task, common clothes were placed in vegetation areas with structured background. Finally, the reflectance signatures of the objects employed in the campaign were collected on the ground by means of a spectroradiometer in order to test spectral matching algorithms in varying atmospheric and illumination conditions. The paper also presents preliminary results on the data in order to give examples of the usefulness of the collected data in discussing the advantages and the critical aspects of target detection techniques in different operating scenarios.
Rossi Alessandro, Acito Nicola, Diani Marco, Corsini Giovanni, De Ceglie Sergio Ugo, Riccobono Aldo, Chiarantini Leandro
Paper for Seminar/Symposium/Conference
SPIE 2014 - International Conference on Remote Sensing (22-25 September 2014, Amsterdam, The Netherlands)
Enter Search Text