This paper proposes a new solution to multi-target joint detection, tracking and classification based on labeled random finite set (RFS) and belief function theory. A class dependent multi-model marginal generalized labeled multi-Bernoulli (MGLMB) filter is developed to analytically calculate the multi-target number, state estimates and model probabilities. In addition, a two-level classifier based on continuous transferable belief model (cTBM) is designed for target classification. To make full use of the kinematic characteristics for classification, both the dynamic modes and states are considered in the classifier, the model dependent class beliefs are computed on the continuous state feature subspace corresponding to different dynamic modes and then fused. As a result that the uncertainty about the classes is well described for decision, the classification results are more reasonable and robust. Moreover, as the estimation and classification problems are jointly solved, the tracking and classification performance are both improved. In the simulation, a scenario contains multi-target with miss detection and dense clutter is used. The performance of multi-target detection, tracking and classification is better than traditional methods based on Bayesian classifier or single model. Simulation results are illustrated to demonstrate the effectiveness and superiority of the proposed algorithm.

Different image fusion systems have been developed to deal with the massive amounts of image data for different applications, such as remote sensing, computer vision, and environment monitoring. However, the generalizability and versatility of these fusion systems remain unknown. This article proposes an efficient regularization framework to achieve different kinds of fusion tasks accounting for the spatiospectral and spatiotemporal variabilities of the fusion process. A joint minimization functional is developed by taking an advantage of a composite regularizer for enforcing joint sparsity in the gradient domain and the frame domain. The proposed composite regularizer is composed of the Hessian Schatten-norm regularization and contourlet-based regularization terms. The resulting problems are solved by the alternating direction method of multipliers (ADMM). The effectiveness of the proposed method is validated in a variety of image fusion experiments: 1) hyperspectral (HS) and panchromatic image fusion; 2) HS and multispectral image fusion; 3) multitemporal image fusion (MIF); and 4) multi-image deblurring. Results show promising performance compared with state-of-the-art fusion methods.

Multibeam forward-looking sonar (MFLS) video object tracking is a challenging problem due to the negative impacts of weak features and background clutter. In this letter, a novel multibeam forward-looking sonar video object tracking method via hybrid regularization scheme is proposed. The proposed regularization scheme is a composite method with truncated $\ell _{1}$ - $\ell _{2}$ sparsity regularization and aberrances repression regularization. While the truncated $\ell _{1}$ - $\ell _{2}$ sparsity regularization explores the structural sparsity of the learned filter to address background clutter, the aberrances repression regularization can alleviate the undesired spatial bounding effect. The resulting optimization problem is solved by alternating direction method of multipliers (ADMM). A proximal operator with truncated soft-thresholding scheme is proposed for the sub-problem with truncated $\ell _{1}$ - $\ell _{2}$ sparsity regularization. Experiments based on five multibeam forward-looking sonar videos for underwater docking validate the effectiveness of the proposed method, compared to other 8 state-of-the-art tracking methods.

Information fusion is a challenging task to extract the required information for high-level applications from various homogeneous or heterogeneous sensors. This paper summarizes some improvements in recent five years in the presented structure of information fusion, including estimation fusion, image fusion and others. Some technologies of information fusion and their variants are discussed. Although the state-of-the-art of the algorithms for information fusion have been proposed, there still remains some fundamental challenges with regard to exploiting the emerging multi-sensors’ characteristics and their special structures. Finally, some potential prospects of estimation fusion and image fusion are discussed.