Urban source detection with mobile sensor networks enhanced with machine learning algorithms

Mobile radiation sensor networks integrated with global positioning system coordinates provide an attractive option for the task of dynamically monitoring a region's background radiation and detecting the illicit movement of radioactive materials. Both statistical and probabilistic approaches have been developed for stationary sensor networks to estimate source parameters. In this paper, a pre-filter framework for mobile sensor networks is presented that estimates multiple sources' positions and intensity without prior knowledge about the sources. The pre-filter framework contains a segmentation algorithm, a clustering algorithm and a maximum likelihood estimation algorithm. With adequate number of measurements, the pre-filter framework estimates source parameters in high accuracy. This offers an efficient alternative to traditional approaches where the multi-source problems are difficult to solve.