Machine learning for localization of radioactive sources via a distributed sensor network

In this paper, we focus on the detection and localization of radioactive sources by exploiting supervised machine learning. Machine learning is utilized in a wide variety of applications due to its effectiveness in prediction and autonomous decision-making. However, applying machine learning would only be effective when representative features for the application can be acquired, through which learning algorithms can be trained. Hence, first, we present a feature extraction technique for radioactive source localization, and then propose a parameter estimation method via machine learning. A distributed sensor network is employed to assist in estimating the radioactive source's location and intensity. We propose a feature extraction method that evaluates a feature vector using the reading and position of each sensor located in a region where a radiation source is detected. The feature extraction is based on a data fusion process, where a single feature value is provided to represent both the reading value and position coordinates corresponding to a given sensor. After the feature extraction, we apply the decision tree machine learning method for regression to localize a radioactive source. To examine the effectiveness of the proposed work, a performance comparison is carried out with recent existing methods in terms of the estimation accuracy and the execution time. Experimental results show that the proposed algorithm provides accurate source intensity estimation and achieves a good compromise between localization accuracy and execution time.