In this work, we propose a novel training paradigm designed to support transfer learning for more effective classification in multispectral airborne imagery. Current state-of-the-art approaches typically rely on either leveraging solely RGB (red-green-blue) pretraining or applying in-domain transfer learning for multispectral imagery classification. Instead, our approach constructs and trains two separate neural network models (backbones): one specifically for wavelengths with available pretrained data (like visible bands) and another trained from scratch on all-bands available in the dataset. These models are then integrated with a fully-connected layer or multi-layered perceptron, which is trained on the features from both networks. This allows us to exploit the significant benefits of generalizable features learned from RGB datasets and the information provided by the full spectrum of multispectral bands. We employ the BigEarthNet and EuroSAT datasets, encompassing Sentinel-2 satellite imagery in the visual and infrared bands. This approach yields considerable performance gains in comparison with other training strategies across every evaluation metric we utilized for these datasets. The results are also consistent across a variety of backbone architectures, underlining the efficacy of our transfer learning technique in the analysis of multispectral data. • A two-stage transfer learning technique is proposed to combine RGB pre-trained models with scratch-trained models on all spectral bands, to integrate features from both to improve classification accuracy and leverage strengths for multispectral airborne imagery. • The strategy has significant performance gains compared to other approached, across multiple neural network backbones and datasets, including BigEarthNet and EuroSAT, highlighting the robustness of the proposed methodology. • The proposed method can be used with state-of-the-art pretraining architectures, enabling the integration and utilization of the latest advancements in machine learning and artificial intelligence.

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.

In this paper, we propose a novel approach to enhance few-shot object detection (FSOD) by utilizing projection modules, which are commonly employed in self-supervised learning to improve feature transferability. We integrate a multilayer perceptron (MLP) as a projection module within a Faster RCNN model and design a tailored training strategy to facilitate the transfer of features from base to novel classes in FSOD. The MLP is used during the base training phase and removed when training on few-shot datasets to leverage more generalized intermediate features for novel classes. Our approach also incorporates affine layers before the classification and regression tasks, which perform element-wise scaling and bias adjustments on the input feature vectors. This effectively reweights the features, enhancing the separability of the feature space for their respective tasks while keeping the parameter count low to minimize overfitting. We evaluate our method against established FSOD techniques, such as Two-Stage Fine-Tuning and Decoupled Faster RCNN, using standard benchmark datasets Pascal VOC and MS COCO. Our results demonstrate significant performance improvements, especially in scenarios with very few training examples.

We address the recovery of block sparse signals by proposing a distributed solution that uses a block-diagonal approximation to the dictionary matrix of the problem. The approximation is found in two stages. First, the Gram matrix of the dictionary matrix is used as a basis for spectral clustering. Afterwards, measurement positions are assigned to the clusters formed from this spectral clustering. The method is then applied to use previous algorithms in the literature of Block Sparse Bayesian Learning in parallel. Moreover, this method also speeds up the algorithm in serial systems. The efficacy of the proposed method is demonstrated in simulations with comparison to the previous Block Sparse Bayesian Learning algorithms.

Quantifying negative information in an efficient way is a challenging task, especially when this information has to be communicated on a network. In this article we leverage the unique properties offered by possibility theory to quantify and approximate the negative information arising in the context of tracking a target with a sensor that has a limited field of view. We also verify experimentally that the corresponding target tracking methodology can be applied in a decentralised manner to a sensor network, while maintaining a performance close to the idealised case where the initial location of the target is better-known.

In this work, we introduce a novel fusion and training strategy aimed at facilitating transfer learning to enhance classification in hyperspectral airborne imagery. Our training strategy has two stages: first we train separate convolutional neural network (CNN) models, one for the bands for which pretraining is available (e.g. visual bands), and a second model trained from scratch on all available wavelengths. These models are then integrated into a new fully-connected layer, which is fine-tuned to fuse the features from both modalities. We use the BigEarthNet and EuroSAT datasets, containing Sentinel-2 satellite imagery in visual and infrared wavelengths. Our approach provides significant performance improvements across all evaluation metrics on the aforementioned data sets, exemplifying the efficacy of our two-stage transfer learning strategy in handling multi-modal data.

We consider the problem of detecting small and manoeuvring objects with staring array radars. Coherent processing and long-time integration are key to addressing the undesirably low signal-to-noise/background conditions in this scenario and are complicated by the object manoeuvres. We propose a Bayesian solution that builds upon a Bernoulli state space model equipped with the likelihood of the radar data cubes through the radar ambiguity function. Likelihood evaluation in this model corresponds to coherent long-time integration. The proposed processing scheme consists of Bernoulli filtering within expectation maximisation iterations that aims at approximately finding complex reflection coefficients. We demonstrate the efficacy of our approach in a simulation example.

Cooperation among multiple autonomous surface and underwater vehicles is an important capability for detection and tracking of underwater objects. Cooperative autonomy in the underwater environment, however, is challenged by the communication bandwidth. In this work, we propose a selective communication scheme that underpins collaborative surveillance under communication constraints. This scheme classifies signal reflections of sonar pulses that are detected by on-board sensor processing as contacts with the object of interest or background using a convolutional neural network. This network is trained using previously labelled contact spectrograms obtained during three sea trials carried out between 2016-2018. The classification scores at the CNN output are ordered to select the few contacts that the underwater modem bandwidth allows for transmission to the network. First, we evaluate the accuracy of the data-driven information selection scheme using recall scores and similar performance measures. Then, we find the accuracy in Bayesian recursive filtering (tracking) of these contacts for different communication rates using established error metrics. The results suggest that the selective scheme yields a favourable surveillance performance communication cost trade-off.

In this work, we propose a data driven trajectory forecasting algorithm that utilizes both recorded historical and streaming trajectory observations. The algorithm performs Bayesian inference on a directed graph the walks on which represent stochastic change point models of trajectory classes. Parameter distributions of these models are learnt from recorded trajectories. Forecasting is then made by calculating the class - or, walk- probabilities and corresponding predictive distributions for a given stream of location and velocity observations. This approach is tailored for the maritime domain and automatic identification system (AIS) data exploitation through the use of an Ornstein-Uhlenbeck process driven stochastic process model that captures vessel motion characteristics. We demonstrate the efficacy of this approach on a real data set.

In this work, we consider the problem of algorithmically predicting rendezvous among vessels based on their trajectory forecasts in a maritime environment. The problem is treated as hypothesis testing on the expected value of the distance between trajectories. We relate this quantity to the first and second degree Wasserstein distances between trajectory forecast distributions. These distributions are obtained using integrated Ornstein-Uhlenbeck process models with the trajectory measurements collected so far. Building upon these results, we propose an algorithm which traverses the trajectories observed so far for detecting rendezvous over a rolling time horizon. We demonstrate the efficacy of the proposed algorithm using simulations.

In this work, we consider the problem of synchronising separately located transmitters and a staring array receiver that also has a local transmitter. The acknowledged benefits of using separate transmitters in active sensing are often undermined by the difficulty in accurate synchronisation of the receiver and the transmitters. In this work, we propose a solution that is based on measurements from non-cooperative objects in the illuminated region. We formulate the problem as parameter estimation in a state space model with individual transmitter channel data cubes as measurements. For maximum likelihood estimation, we use an expectation maximisation type iterative bound optimisation using distributions found by track-before-detect together with explicit formulae derived here for the related score function for chirp waveforms. We demonstrate that the proposed approach is capable of achieving very high accuracy with errors on the order of small fractions of the pulse width thereby enabling coherent processing in bi-static channels.

Motivated by object tracking applications with networked sensors, we consider multi sensor state space models. Estimation of latent parameters in these models requires centralisation because the parameter likelihood depend on the measurement histories of all of the sensors. Consequently, joint processing of multiple histories pose difficulties in scaling with the number of sensors. We propose an approximation with a node-wise separable structure thereby removing the need for centralisation in likelihood computations. When leveraged with Markov random field models and message passing algorithms for inference, these likelihoods facilitate decentralised estimation in tracking networks as well as scalable computation schemes in centralised settings. We establish the connection between the approximation quality of the proposed separable likelihoods and the accuracy of state estimation based on individual sensor histories. We demonstrate this approach in a sensor network self-localisation example.