Recently, transformer-based Unmanned Aerial Vehicle (UAV) trackers have achieved notable success. However, the computationally intensive transformer model limits these trackers to static templates and shallow backbone networks, hampering their discriminative power and localization precision. Here, we propose a novel discriminative Siamese deep-coupling framework. This framework constructs a lightweight fine-grid anchor-free Siamese tracker with high spatial resolution specifically tailored for UAV scenarios, and complements its discriminative power with a targeted online discriminator. To achieve this, an efficient distractor detector is developed via knowledge transfer, enabling targeted detection of distractors that disturb the Siamese tracker. These distractors are utilized as training samples to construct a targeted online discriminator, which is deeply coupled with the Siamese tracker to enhance its discriminative power and specifically suppress hard distractors that hinder tracking performance. Additionally, a leading principal submatrix cluster sample space model and a scene-aware dynamic update strategy are developed to purify online samples and dynamically schedule the online discriminator update, significantly reducing the computational cost of the online discriminator optimization and boosting the tracker’s real-time performance. Finally, extensive experiments on eight UAV tracking benchmarks demonstrate that our tracker surpasses state-of-the-art transformer-based UAV trackers while achieving 70 FPS on CPU.

Existing correlation filter (CF) tracking methods are fragile for boundary effects, vague target information, and heuristic model updating, as these limitations degrade the detection ability of the learned filter. In response to that, this article embarks on basic CF learning and presents a novel distractor-aware template-coupled correlation filter (DATC-CF) by exploiting the spatial-temporal appearance context of the target, which aims at improving the discriminative ability of the learned filter against distractive background and the descriptive ability in adapting unexpected scenes. Specifically, the power of spatial context comes from a distractor-aware regularizer weighted by background distractors. By adaptively optimizing the weight of each distractor, our filter training can focus more on the critical distractors. The temporal context is represented by a dynamic template set, and we formulate a template-coupled regularizer that can make use of the commonality over all templates while maintaining a passive filter update under a multi-template learning scheme. DATC-CF integrates the two regularizers and is summarized as a multi-variable joint optimization problem where a filter ensemble can be learned. With DATC-CF, a multi-model tracking framework DATC_MM is developed by maximizing the posterior distribution over the learned filters. For robust tracking, we further apply high-confidence updating and establish a complementary distractor-aware color detector to restore the CF tracking failures. Finally, experiments on several large-scale benchmark datasets demonstrate the effectiveness of the proposed tracking methods against state-of-the-art trackers.

In the process of microseismic monitoring, the accuracy of seismic phase picking has a great influence on the precision of seismic localization. Recently, neural network models have achieved satisfying results in seismic phase picking. But these models are mainly applicable to natural earthquakes. In this letter, we present a fully convolutional network for microseismic phase picking called MSNet. The performance of the model is improved from three aspects: model structure, training data, and training strategy. MSNet has two decoders to allow for auxiliary learning, and its structure is improved with reference to state-of-the-art fully convolutional network. Our training data derives from the Stanford Earthquake Dataset (STEAD). It is selected and processed to meet our requirements. In the training process, a special early stopping strategy is used to obtain the model with best generalization performance. The model is tested on fracture-induced microseismic monitoring data from a coal mine in Shaanxi Province, China. Methods for comparison include short- and long-time average ratio, the original PhaseNet and a retrained PhaseNet. The number of correct P picks of MSNet is 18.7% more than PhaseNet, and MSNet achieves a 16.2% higher precision than PhaseNet on S phase. The event locating results verified the accuracy of MSNet. Our method provides an idea to build a microseismic phase picking model with better generalization performance and lower labor cost.

Abstract Visual object tracking is a complicated problem due to varied challenges such as occlusion, background clutter, target appearance changes, abrupt target motion, and illumination variations. Different tracking algorithms perform well on different challenges because of their unique strengths. In this paper, a dual‐integration framework that integrates the strengths of multiple models while avoiding their weaknesses is proposed. In the proposed framework, several evaluation criteria and multiple models updated via different processes are combined. To distinguish the target from the distractors caused by different criteria, the motion dynamic model and the forward‐backward analysis is introduced to provide information that is complementary to appearance information. In addition, for discriminating occlusion, a spatial‐temporal occlusion‐aware approach is further proposed. The detected occlusion results are applied in avoiding contamination of the appearance model. Extensive experiments on multiple benchmarks demonstrate that this proposed method improves the overlap rate of a hand‐crafted feature‐based tracker with relative gains of 4%, 1.6%, 1.9%, 2.2% and 8.5% on OTB‐2015, OTB‐2013, Temple‐Color, UAV123 and UAV20L, respectively. Also, the experimental results demonstrate that our approach outperforms a deep feature‐based tracker in overlap rate by 1.4%, 1.6%, 2.3%, 2.5% and 3.5% on OTB‐2015, OTB‐2013, Temple‐Color, UAV123 and UAV20L, respectively.

Robust and accurate tracking for fast moving object in correlation filter framework is a challenging research problem. The appearance model update strategy impact on performance is usually very significant and hence is worth studying. Unfortunately, very few works focus on this component. This study proposes an adaptive appearance model update method, which utilises both average peak‐to‐correlation energy (APCE) threshold and gradient APCE threshold to measure tracking reliability. When tracking is unreliable, the update rate is modified and the initial template information is used to assist in correcting the model parameters. Different from conventional methods, both APCE threshold and gradient APCE threshold are enhanced by weighting the coefficients. More importantly, the gradient APCE threshold can capture the rapid decline of APCE which implies that the object is in fast motion. Sufficient evaluations on OTB datasets demonstrate that the proposed method can tackle challenging videos well, especially the authors obtain a gain of 4.6% in precision score on the selected 32 videos with fast motion attribute, and a mean gain of 3.0 and 2.2% in precision and success plots on OTB50, compared with the baseline tracker SAMF.

To address the challenge of repetitive target appearance variation and frequent occlusion, existing visual tracking methods either handle corrupted samples or correct the appearance model. In this study, the authors propose a novel framework that successfully combines these two strategies. In their method, the base tracker is an improved discriminative correlation filter‐based tracker, in which an independent classifier is employed to alleviate the problem of corrupted samples; the best model is selected for improvement from a group of models, which they call a ‘model colony’. The model colony is composed of models updated via different processes. The correlation output and the peak‐to‐sidelobe ratio are used to evaluate each model in the model colony. In addition, they propose a novel criterion called the maximum‐to‐others ratio for superior model selection. Experiments on 80 challenging sequences show that their tracker outperforms state‐of‐the‐art trackers. In addition, experimental results demonstrate that their formulation significantly improves the performance of their base tracker.

Visual object tracking is an important and challenging task in computer vision. In this study, the authors propose a novel visual tracking approach by decomposing the tracking task into translation and scale estimation. In translation estimation, they employ multiple adaptive correlation filters with features of hierarchical convolutional neural networks (CNNs) to more accurately estimate the target location. To make full use of multi‐level features from different CNN layers, they propose an adaptive weighted algorithm to fuse correlation response maps. In scale estimation, a one‐dimensional correlation filter with histogram of oriented gradient (HOG) features is employed to estimate the scale variation. Extensive experimental results on 50 challenging benchmark video sequences demonstrate that the proposed algorithm outperforms state‐of‐the‐art algorithms.