Recently, deep convolutional neural networks (DCNN) have been widely used in semantic segmentation tasks and have achieved high segmentation accuracy. However, most algorithms based on DCNN have high computational complexity, making them unsuitable for real‐time segmentation. To solve this problem, this paper proposes a real‐time semantic segmentation algorithm based on the STDC network. The algorithm adopts an “encoder–decoder” embedded in a U‐shaped architecture to realize real‐time segmentation while maintaining high accuracy. Following the encoder, a mixed pooling attention module is designed to expand the receptive field, enhancing the network model’s learning ability in complex scenarios. Then, a feature fusion module is used for combining features from different stages, and channel attention based on atrous convolution is employed to expand the receptive field and avoid dimensionality reduction learning. Finally, a Tversky‐based detail loss function is used to encode more spatial details. The proposed algorithm was extensively tested on the challenging Cityscapes and CamVid datasets, and the experimental results showed that the proposed algorithm obtained 76.4% and 72.8% of mIoU, respectively. Meanwhile, our algorithm achieves 105.2 FPS and 165.6 FPS inference speed with a single NVIDIA GTX 1080Ti GPU, meeting the real‐time segmentation requirements. The proposed algorithm can conduct real‐time segmentation while maintaining high accuracy, achieving a good balance between accuracy and speed.

Discriminative correlation filter (DCF)-based UAV tracking algorithms have drawn much attention due to their outstanding robustness and high computational efficiency. However, these algorithms are easily disturbed by background noise and abrupt changes in target appearance, leading to tracking failure. To address the issues above, we propose a real-time UAV object tracking algorithm with adaptive spatial-temporal attention. Specifically, we construct two filters with different roles based on the training sample's target foreground and environmental background. The spatial attention filter is implemented by incorporating a spatial context regularizer into the traditional DCF paradigm, which fully utilizes background environmental information to suppress background environmental noise and effectively distinguish between the target and the background. The temporal attention attention filter focuses on the continuity of the target samples, modeling only the target patch samples during the training process and introducing a temporal context regularizer, which substantially enhances the tracker's robustness against target occlusions and deformations. The two are jointly optimized by the Alternating Direction Method of Multipliers (ADMM) algorithm, which is mutually constrained during training and complemented during detection. Extensive experiments on three mainstream UAV benchmarks demonstrate the tracking advantages of the proposed algorithm.

Although the discriminative correlation filter‐ (DCF)‐based tracker improves tracking performance, some object representation issues can still be further optimized. On the one hand, the DCF tracker’s deep convolutional features contain many noisy channels, and assigning the same weights to multiple channels cannot distinguish the importance of different channels. On the other hand, a simple weighted fusion approach cannot fully utilize the benefits of different feature types. We propose a visual object tracking algorithm based on adaptive channel weighting and feature game fusion to solve these problems. In this study, an adaptive channel weighting strategy is designed to assign suitable weights to each channel based on the average energy ratio of the target and background regions in the feature channels and prune the channels with low weights to improve feature robustness and reduce computational complexity. Simultaneously, the game theory concept is introduced in the multifeature fusion. The handcrafted features are combined with shallow and deep convolutional features according to feature complementarity. Then, the two combined features are seen as two sides of the game, continuously gamed during the tracking process to generate a feature model with a higher representation capacity. Extensive experiments are conducted on four mainstream visual tracking benchmark datasets, including OTB2015, VOT2018, LaSOT, and UAV123. The experimental results show that the proposed algorithm performs outstandingly compared to the state‐of‐the‐art trackers.

Discriminative correlation filters (DCFs) have shown promising tracking performance in recent years thanks to the powerful representation ability of deep features. However, a large number of target-irrelevant channels in deep features limits the tracking performance and increases the computational cost. To eliminate the negative impact of noisy channels and improve the utilization efficiency of deep features in DCF-based trackers, we present an adaptive feature channel selection method for robust visual tracking. Our method adaptively chooses the most discriminative channels to learn a more robust target appearance model, which is achieved by evaluating the energy relationship between background and foreground in each feature channel. Moreover, according to the feedback of channel selection, an adaptive model update strategy is proposed to alleviate the model degradation problem caused by incorrect model updating. Extensive experimental results obtained on five popular tracking benchmarks demonstrate the effectiveness of the proposed algorithm and its superiority over the state-of-the-art trackers.

Community detection is a critical task for complex network analysis. It helps us to understand the properties of the system that a complex network represents and has significance to a wide range of applications. Nowadays, the challenges faced by community detection algorithms include overlapping community structure detection, large scale network analysis, dynamic changing of analyzed network topology and many more. In this paper a self-organizing community detection algorithm, based on the idea of swarm intelligence, was proposed and its parallel algorithm was designed on Giraph++ which is a semi-asynchronous parallel graph computation framework running on distributed environment. In the algorithm, a network of large size is firstly divided into a number of small sub-networks. Then, each sub-network is modeled as a self-evolving swarm intelligence sub-system, while each vertex within the sub-network acts iteratively to join into or leave from communities based on a set of predefined vertex action rules. Meanwhile, the local communities of a sub-network are sent to other sub-networks to make their members have a chance to join into, therefore connecting these self-evolving swarm intelligence sub-systems together as a whole, large and evolving, system. The vertex actions during evolution of a sub-network are sent as well to keep multiple community replicas being consistent. Thus network communication efficiency has a great impact on the algorithm' performance. While there is no vertex changing in its belonging communities anymore, an optimal community structure of the whole network will have emerged as a result. In the algorithm it is natural that a vertex can join into multiple communities simultaneously, thus can be used for overlapping community detection. The algorithm deals with vertex and edge adding or deleting in the same way as the algorithm running, therefore inherently supports dynamic network analysis. The algorithm can be used for the analysis of large scale networks with its parallel version running on distributed environment. A variety of experiments conducted on synthesized networks have shown that the proposed algorithm can effectively detect community structures and its performance is much better than certain popular community detection algorithms.

Community structure analysis is a critical task for complex network analysis. It helps us to understand the properties of the system that a complex network represents, and has significance to a wide range of real applications. The Label Propagation Algorithm (LPA) is currently the most popular community structure analysis algorithm due to its near linear time complexity. However, the performance of the LPA has proven to be unstable and the correctness of community assignment of nodes is unsatisfactory. In this paper a Self-Organizing Community Detection and Analytic Algorithm (SOCDA2) based on swarm intelligence is proposed. In the algorithm, a network is modeled as a swarm intelligence system, while each node within the network acts iteratively to join or leave communities based on a set of pre-defined node action rules, in order to improve the quality of the communities. When there is not a node changing its belonging community anymore, an optimal community structure will emerge as a result. A variety of experiments conducted on both synthesized and real-world networks have shown results which indicate that the proposed algorithm can effectively detect community structures and the performance is better than that of the LPA. In addition, the algorithm can be extended for overlapping community detection and be parallelized for large-scale network analysis.