Search

For infrared target segmentation, convolutional neural network-based (CNN-based) methods have largely improved the segmentation accuracy. Previous CNN-based methods typically improved the segmentation accuracy by enhancing and aggregating multi-level features. However, the existence of a semantic gap between features from different semantic levels limits the improvement of feature aggregation. To address this problem and improve the segmentation accuracy of infrared targets with tiny sizes and irregular shapes, the multiattention shape adaptive network (MASANet) is proposed in this article. Specifically, a spatial feature fusion dense-nested Unet (SFFDUnet) is designed as the feature extraction and aggregation backbone. Based on SFFDUnet, the bi-directional shape adaptive attention with channel-wise shape enhance (BSAA-CSE) block is proposed as a multiattention mechanism to reduce the semantic gap between features. Moreover, the multiscale multibranch deep supervision is introduced to stabilize the training process and enable model pruning. In addition, a new infrared target dataset CITY-IRTD is also constructed to evaluate the performance of our method. Experiments demonstrate the effectiveness of our method in reducing the semantic gap between features to be fused. As a result, our MASANet has achieved better performance on both public and our self-developed datasets, compared to other state-of-the-art infrared target segmentation methods. Codes and the dataset will be released at https://github.com/twentyfiveYang/MASANet_CITY-IRTD.

An efficient anomaly detection and localisation mechanism is crucial for achieving high-quality network services. In particular, learning-based methods have recently been developed to achieve this goal by discovering helpful information from a massive amount of heterogeneous network data. However, heterogeneous data from various network components lead to significant challenges and an unexpected burden for analysis. The distributed scale of networking systems challenges data integrity and knowledge retrieval due to the separation of coupled functions over the distributed system. In this article, an insightful survey is performed by thoroughly reviewing recent academic and industrial contributions regarding anomaly detection and localisation. To tackle the issues, we propose a new framework to effectively learn informative representations of heterogeneous data and fuse this information for efficient anomaly detection and localisation. Furthermore, a case study is presented for anomaly detection and localisation through learning data representations and performing heterogeneous data fusion.

Hematopoietic stem cells (HSCs), which are localized in the bone marrow of adult mammals, come from hematopoietic endothelium during embryonic stages. Although the basic processes of HSC generation and differentiation have been described in the past, the epigenetic regulation of embryonic hematopoiesis remains to be fully described. Here, by utilizing an in vitrodifferentiation system of mouse embryonic stem cells (ESCs), we identified more than 20 microRNAs that were highly enriched in embryonic hematopoietic cells, including some (e.g. miR-10b, miR-15b, and miR-27a) with previously unknown functions in blood formation. Luciferase and gene expression assays further revealed combinational binding and regulation of these microRNAs by key transcription factors in blood cells. Finally, bioinformatics and functional analyses supported an interactive regulatory control between transcription factors and microRNAs in hematopoiesis.

This study is mainly focusing on the problem of spacecraft close-range proximity with obstacle avoidance in the presence of complex shape. A novel Gaussian mixture model–based nonsingular terminal sliding mode control (GMM-NTSMC) is proposed. This is achieved by developing GMM-based potential function with a switching surface of NTSMC. It is theoretically proved that the closed-loop system is globally stable. The main contribution of this study is that the GMM-based avoiding strategies, which include the GMM-based terminal sliding mode control (GMM-TSMC) and GMM-NTSMC, can solve the collision avoidance problem considering complex shape while the artificial potential function–based terminal sliding model control (APF-TSMC) fails. Moreover, the GMM-NTSMC and the GMM-TSMC require less energy with respect to the APF-TSMC. Furthermore, the GMM-NTSMC retains the advantage of the NTSMC and can avoid singularity problem while GMM-TSMC cannot. Finally, numerical simulations are performed to verify the effectiveness and superiority of the proposed GMM-NTSMC.