Broadband passive SONAR systems must detect multiple acoustic sources in environments marked by high noise and limited prior information. Traditional model-based approaches, such as Cell Averaging Constant False Alarm Rate (CA-CFAR), rely on analytical models, which can fail to fully capture complex conditions, creating a need for data-driven methods. We propose a detection framework that integrates a Poisson Random Finite Set (RFS)-based loss function with a Vision Transformer (ViT) architecture. The ViT component processes Bearing Time History (BTH) waterfall data patches, capturing local and global acoustic features, while the Poisson RFS loss naturally accommodates an unknown number of sources. A test-time augmentation (TTA) strategy further boosts performance by exploiting the circular symmetry in bearing data. Experimental results show that our approach improves upon a conventional CFAR detector and CNN-based baselines across Receiver Operating Characteristic (ROC), Precision-Recall, and detection probability metrics. In particular, pairing ViT with the RFS loss yields higher accuracy and robustness to noise, all within a computationally feasible framework for real-time detection tasks.

This paper presents an extended target tracking method to track trains using Distributed Acoustic Sensing (DAS) data. The problem is approached using a measurement likelihood based on a Set of Points on a Rigid Body (SPRB) model applied to a clustered version of the Poisson Multi-Bernoulli Mixture filter. The method efficiently handles asymmetric noise within the set of measurements returned by each train, and proposes a solution to merged measurements appearing at crossings. We use experimental data obtained from trains to show that the proposed algorithm has lower localisation and false target error, leading to better performance in terms generalized optimal sub-pattern assignment (GOSPA) metric.

Under the standard multiple target tracking models and a Poisson point process birth model, the Poisson multi-Bernoulli mixture (PMBM) filter provides the closed-form recursion to computing the posterior density over the set of targets. Without approximations, the PMBM computational complexity rapidly rises in time due to the increasing number of data association hypotheses. This paper presents innovative strategies for merging Bernoulli components for the same potential target reducing the number of single-target hypotheses in the PMBM filter, aiming to lower its computational complexity while keeping its performance high. We use several measures to compute the similarity between different Bernoulli components. Simulation results show that the proposed algorithms show performance close to the PMBM filter without Bernoulli merging, as measured by the generalized optimal sub-pattern assignment (GOSPA) metric, with a significantly reduced execution time.