Few-shot object segmentation with a new feature aggregation module

The success of convolutional neural network for object segmentation depends on a large amount of training data and high-quality samples. But annotating such high-quality training data for pixel-wise segmentation is labor-intensive. To reduce the massive labor work, few-shot learning has been introduced to segment objects, which uses a few samples for training without compromising the performance. However, the current few-shot models are biased towards the seen classes rather than being class-irrelevant due to lack of global context prior attention. Therefore, this study aims at proposing a few-shot object segmentation model with a new feature aggregation module. Specifically, the proposed work develops a detail-aware module to enhance the discrimination of details with diversified attributes. To enhance the semantics of each pixel, we propose a global attention module to aggregate detailed features containing semantic information. Furthermore, to improve the performance of the proposed model, the model uses support samples that represents class-specific prototype obtained by respective category prototype block. Next, the proposed model predicts label of each pixel of query sample by estimating the distance between the pixel and prototypes. Experiments on standard datasets demonstrate significance of the proposed model over SOTA in terms of segmentation with a few training samples.     

Ship detection based on multi-scale weighted fusion

Ship surveillance plays an important role in ensuring the safety of maritime transportation and navigation. Due to the influence of factors such as waves and special weather, the existing detection methods still cannot balance the accuracy, speed and the parameters of the model in the changeable and complex marine environment. To solve this problem, this paper proposes an improved real-time method based on YOLOv5, which has few parameters and achieves high detection accuracy with little memory and computation cost. Collaborative Attention (CA) mechanism is added to the network structure, which enables the model to more accurately locate and identify target regions. We also design a Spatial Pyramid Pooling module (SPP) and a weighted pyramid network called Bidirectional Feature Pyramid Network (BiFPN) based on the characteristics of the ships to better fuse feature information. Transformer encoder is introduced to capture long-distance dependencies and preserve global and local features to the greatest extent. Furthermore, the ability of our proposed structure to localize objects at each stage is improved through integrating the output of multiple modules. The experimental results show that, the comprehensive performance of this method is better than the existing technology in ship detection on different evaluation criteria.     

Ultraefficient Non-Doped Deep Blue Fluorescent OLED: Achieving a High EQE of 10.17% at 1000 cd m−2 with CIEy<0.08

The pursuit for efficient deep blue material is an ever-increasing issue in organic optoelectronics field. It is a long-standing challenge to achieve high external quantum efficiency (EQE) exceed 10% at brightness of 1000 cd m⁻² with a Commission International de L'Eclairage (CIE_y) <0.08 in non-doped organic light-emitting diodes (OLEDs). Herein, this study reports a deep blue luminogen, PPITPh, by bonding phenanthro[9,10-d]imidazole moiety with m-terphenyl group via benzene bridge. The non-doped OLED based on PPITPh exhibits an exceptionally high EQE of 11.83% with a CIE coordinate of (0.15, 0.07). The EQE still maintains 10.17% at the brightness of 1000 cd m⁻², and even at a brightness as high as 10000 cd m⁻², an EQE of 7.5% is still remained, representing the record-high result among non-doped deep-blue OLEDs at 1000 cd m⁻². The unprecedented device performance is attributed to the reversed intersystem crossing process through hot exciton mechanism. Besides, the maximum EQE of orange phosphorescent OLED with PPITPh as host is 32.02%, and remains 31.17% at the brightness of 1000 cd m⁻². Such minimal efficiency roll-off demonstrates that PPITPh is also an excellent phosphorescent host material. The result offers a new design strategy for the enrichment of high-efficiency deep blue luminogen.     

基于关系强度理论与反馈机制的信息传播动态网络表示

现有多数网络表示学习方法不能很好地贴合真实世界的信息传播网络,且无法对信息传播动态网络的时间特性与动力学演化特征进行有效建模。提出一种新的信息传播动态网络表示模型,基于关系强度将信息传播动态网络划分为关系网络与传播网络,并分别计算变化节点对的概率密度和邻接矩阵。通过更新节点注意力强度矩阵,聚合节点邻域变化信息,并融合节点邻域变化信息、自身历史信息以及外部影响因素,对信息传播动态网络进行归纳式表示学习。引入反馈机制,将最新的节点表示反馈到邻居节点,解决网络表示不及时的问题,提升网络表示性能。实验结果表明,与Know-Evolve、DyRep、LDG等模型相比,该模型的命中率和平均排名提升显著,与LDG模型相比,其时间效率在Social Evolution数据集和Github数据集上分别提升了91.8%、87.2%。     

基于人脸关键特征提取的表情识别

自然场景下人脸表情由于受遮挡、光照等因素影响,以及表情局部变化细微,导致现有人脸表情识别方法准确率较低。提出一种人脸表情识别的新方法,以ResNet18为主干网络,利用残差连接模块加深网络结构,以提取更多深层次的表情特征。通过引入裁剪掩码模块,在训练集图像上的某个区域进行掩码,向训练模型中增加遮挡等非线性因素,提升模型在遮挡情形下的鲁棒性。分别从特征图的通道和空间两个维度提取表情的关键特征,并分配更多的权重给表情变化明显的特征图,同时抑制非表情特征。在特征图输出前加入Dropout正则化策略,通过在训练中随机失活部分神经元,达到集成多个网络模型的训练效果,提升模型泛化能力。实验结果表明,与L2-SVMs、IcRL、DLP-CNN等方法相比,该方法有效提高了表情识别准确率,在2个公开表情数据集Fer2013和RAF-DB上的识别准确率分别为74.366%和86.115%。     

Pre-Protonated Vanadium Hexacyanoferrate for High Energy-Power and Anti-Freezing Proton Batteries

Proton batteries have been considered as an innovative energy storage technology owing to their high safety and cost-effectiveness. However, the development of fast-charging proton batteries with high energy/power density is greatly limited by feasible material selection. Here, the pre-protonated vanadium hexacyanoferrate (H-VHCF) is developed as a proton cathode material to alleviate the capacity loss of proton-free electrode materials during electrochemical tests. The pre-protonation process realizes fast and long-distance transport of protons by shortening diffusion path and reducing migration barriers. Benefitting from the enhanced hydrogen bonding network combined with dual redox reactions of V and Fe in protonated H-VHCF cathode, a high energy density of 74 Wh kg⁻¹ at 1.1 kW kg⁻¹, and a maximum power density of 54 kW kg⁻¹ at 65 Wh kg⁻¹ is achieved for the asymmetric proton batteries coupling with MoO₃/MXene anode. Proton transport and double oxidation-reduction center are verified by theoretical calculations and ex situ experimental measurements. Considering the anti-freezing availability of proton batteries, 82.5% of its initial capacity is maintained after 10000 cycles under −40 °C at 0.5 A g⁻¹. As a proof-of-concept, flexible device fabricated by optimized electrodes and hydrogel electrolytes can power up a light-emitting diode even under a bent state.     

基于点对群网络反馈机制的恶意代码传播模型

在点对群信息共享网络中,群体成员之间交流频繁并且同一群体内的成员可以同时从信息源接收到相同信息,依据点对群网络的这2个特点,考虑从恶意代码感染中恢复后的节点作用,在点对群网络中建立一种具有动态反馈防治信息功能的易感-感染-反馈-免疫(SIFR)模型。在经典易感-感染-免疫(SIR)传播模型的基础上引入反馈节点,通过动态共享防治信息遏制恶意代码在点对群网络中的传播。根据计算得到SIFR模型的平衡点和传播阈值,构建相应的Lyapunov函数,证明了平衡点的局部和全局稳定性。数值模拟实验结果显示：当反馈率取0.000 1时,SIFR模型相较于经典SIR模型在传播阈值小于1的情况下,感染节点在峰值处的数量降低了36.16%,能更早更快地趋近于0;当传播阈值大于1时,同一时间的感染节点数量有所减少,趋于稳定的感染节点数量降低了80%。上述实验结果表明SIFR模型应用在点对群网络中能够更好地遏制恶意代码的传播,且反馈率越高,遏制效果越好。     

轻量级区块链技术综述

传统的区块链技术为了保证交易账本的全网共识和不可篡改性,要求矿工节点具有强大的计算能力和足够的存储空间,这就限制了资源受限的设备加入区块链.近几年,区块链技术已经拓展到金融经济、医疗健康、物联网、供应链等多个领域,但是这些应用场景存在大量算力弱、存储容量低的设备,这给区块链的应用带来了巨大挑战.为此轻量级的区块链技术应运而生.从轻量级计算和轻量级存储两方面出发,总结当前轻量级区块链的研究现状,对比分析各个方案的优缺点.最后展望未来轻量级区块链的发展.     

联邦学习贡献评估综述

数据不动的联邦学习框架是多个数据持有方合作训练机器学习模型的新范式.多个数据持有方参与联邦学习时的贡献评估是联邦学习的核心问题之一.参与方贡献评估需要兼顾有效性、公平性和合理性等要素,在理论方法与实际应用中均面临多项挑战.贡献评估首先需要明确如何度量数据价值,然而数据估值存在主观性与依赖于实际任务场景的特点,如何设计有效、可靠并对恶意数据鲁棒的数据估值指标是第一大挑战.其次,联邦学习合作中的参与方贡献评估是经典的合作博弈问题,如何制定公平合理的参与方贡献评估方案,实现参与方一致认可的博弈平衡是第二大挑战.最后,参与方贡献评估往往计算复杂度高,同时,联邦学习中围绕模型的数据估值时间开销大,因此,在实践中如何设计高效且准确的近似算法是第三大挑战.近年来,为了有效地解决上述挑战,学术界对联邦学习中的贡献评估问题展开了广泛的研究.首先,简要介绍联邦学习与参与方贡献评估的背景知识;然后,综述数据估值指标、参与方贡献评估方案和相关优化技术;最后,讨论了联邦学习贡献评估仍面临的挑战并展望未来研究的发展方向.     

多机制优化的模拟试验台数据流实时迁移研究

航空发动机高空模拟试验作为检验航空发动机性能的重要手段,其试验流程中采集的各传感器测点数据极为重要与珍贵。为了使数据信息的展示更为直观并且更深入挖掘数据潜在的价值,提出基于多框架集成的航空发动机高空模拟试验台数据流实时迁移及可视化技术。首先,利用MyBatis-Plus自动视图层构建可视化技术将存储于底层历史数据库中的信息直观地展示于Vue框架的浏览器端表结构空间中,方便试验员查看分析;然后利用基于GateWay限流机制的JDBC接口融合全量/增量和基于负载均衡机制的MyBatis框架融合多线程并行传输流的数据流迁移技术,将历史数据库中可供后续算法分析利用的重要数据信息同步至本地数据库中;最后,利用Vue框架的数据流前端监控界面,验证数据可视化展示的直观性、完整性、有效性和数据流同步的实时性。