Assessment of intelligent sensor systems - Final report

The task for the project "Assessment of intelligent sensor systems" was to further develop methodology to assess sensors systems that uses artificial intelligence (AI), to be able to compare the value of AI with traditional sensor systems. When performing assessment of sensor systems a good method is required that takes into account all factors such as: the sensor system, tactics, environment, economics, human system interaction, education etc. In this study, we used the assessment method COAT/TVS and further developed this as well as tools for assessment of sensor systems that use artificial intelligence (AI). To evaluate the enhanced assessment methodology and tools, an application example was studied. The example is assessment of the ability of an small Remotely Piloted Aircraft (RPA) system doing reconnaissance against a hostile motor rifle brigade. To be able to discuss methods and tools, only with unclassified information was used. The RPA and air defense system that have been assessed may differ slightly from actual sensor systems, as data in some cases have been unknown. In these cases, reasonable values have been assumed. The target acquisition radar on the air defense system 2S6 Tunguska has been simulated to analyze the ability of the radar to detect RPA. The simulations show that it is difficult, but not impossible, to optimize a flight path for reconnaissance near an air defense system in a known location so that detection can be avoided at all times. When several air defense systems work together, it becomes almost impossible for an RPA to avoid being detected. An assessment of the ability of the self-propelled anti-aircraft weapon to detect small RPAs was carried out and, for example, shows the probability of an RPA being followed and being shot down as a function of time in the mission. The project has studied the ability that an automatic target recognition algorithm can add to an small RPA such as UAV 05B Korpen when reconnaissance with a thermal IR camera against various vehicles. In the project, an experiment has been carried out to compare human performance with the ability of a target recognition algorithm to detect and recognize vehicles. The results show that the trained target recognition algorithm today performs at the same level as a human who must respond within five seconds. The project has participated in a study of automatic target recognition with micro-Doppler radar and deep learning. The study used various objects such as a two small helicopters, a small model aircraft with a fixed wing, one quadcopter and some birds. The result is very promising, but is based on a limited amount of data. More work and data are required to develop an algorithm that is sufficiently robust and can handle a sufficient scope of conditions to be practically useful.