基于RetinaNet-CPN网络的视频人体关键点检测

为解决基于视频流的人体关键点检测效果不佳及视频流切片后可能会发生运动模糊的问题,提出了一种改进的RetinaNet-CPN网络对人体关键点进行检测,有效解决切片后运动模糊图像的干扰并提高了人体关键点的检测准确率.视频流切片后,先用改进的RetinaNet网络检测出图片中的所有人并对每个目标框做模糊检测,对大于阈值的目标框做去模糊处理,最后用引入注意力机制的CPN网络提取关键点.将RetinaNet衡量预测框与真实框差异的IOU函数改成DIOU后,在仿真实验中目标检测AP提高了近3％;对于模糊的图片,利用匀速直线运动频谱特征估算出的模糊核与实际模糊核相差不大,对其做去模糊处理后基本能恢复出原清晰图片;同时引入注意力机制为各通道和特征层分配合理的权重,使得CPN检测AP提高近1％,AR提升0.5％.     

融入类贡献抑制因子的灰度级模糊C均值图像分割

基于灰度级模糊C均值图像分割算法具有分割速度快的优势。由于无损检测图像中背景类和目标类差异较大,该算法不能有效地将目标分割出来,故提出改进的基于灰度级的模糊C均值算法。构建了一种与类大小反向相关的类贡献抑制因子表达式,将之融入目标函数后能够降低较大类对目标函数的贡献,这可避免较小类的聚类中心受较大类的影响而靠近较大类的聚类中心。最小化新的目标函数可得新形式的隶属度和聚类中心表征形式。采用类大小差异较大的无损检测图像进行试验,结果显示本文算法得到的分割图像视觉效果良好,而且指标G_mean也更高,进一步提升了基于灰度级模糊C均值算法适应能力。     

基于神经网络与模糊算法图像边缘检测方法*

针对传统图像边缘检测方法中出现毛边、噪边、边缘定位不精确等缺点,提出一种神经网络与模糊算法相结合的检测方法。根据图像特征,将图像分为高频和低频部分分别处理,高频部分适宜用双层网络结构,可以很好地减弱噪声;对于图像低频部分,将模糊理论引入到边缘检测中,能够检测出弱边。最后对检测出的两个图像边缘进行融合,实验结果证明得出的检测效果较好,比传统边缘检测算子所获结果有很大改善。     

基于模糊层次分析的动态联盟伙伴选择策略

论述了基于模糊层次分析(AHP)的动态联盟伙伴选择策略。综合运用AHP方法和模糊决策方法构建了动态联盟伙伴选择的排序模型 ,通过求解该模型来得到参考排序值 ,力求使排序的结果能够客观、公正地反映不精确决策所蕴含的客观规律 ,以保证选择策略的准确性     

真空水冷铜坩埚感应炉模糊控制专家系统的研究

研究真空水冷铜坩埚感应炉中模糊控制专家系统。阐述该炉控制系统复杂性 ,介绍模糊控制专家系统的主要结构和各主要部分的功能和一些算法     

机械零件刚度的模糊可靠性设计

介绍了机械零件刚度模糊可靠性设计变形分布参数和刚度隶属函数的确定,导出了零件刚度模糊可靠度的计算公式,并通过机床主轴模糊可靠性设计实例说明机械零件模糊可靠性的计算方法.     

Beyond multi-adjoint logic programming

In ‘multi-adjoint logic programming’, MALP in brief, each fuzzy logic program is associated with its own ‘multi-adjoint lattice’ for modelling truth degrees beyond the simpler case of true and false, where a large set of fuzzy connectives can be defined. On this wide repertoire, it is crucial to connect each implication symbol with a proper conjunction thus conforming constructs of the form (←_i, &_i) called ‘adjoint pairs’, whose use directly affects both declarative and operational semantics of the MALP framework. In this work, we firstly show how the strong dependence of adjoint pairs can be largely weakened for an interesting ‘sub-class’ of MALP programs. Then, we reason in a similar way till conceiving a ‘super-class’ of fuzzy logic programs beyond MALP, which definitively drops out the need for using adjoint pairs, since the new semantics behaviour relies on much more relaxed lattices than multi-adjoint ones.     

基于T-S 模糊模型的采样数据网络控制系统H∞ 输出跟踪控制

研究具有采样数据的基于T-S (Takagi-Sugeno) 模糊模型网络控制系统H输出跟踪控制问题. 提出将采集器端数据采样周期、数据传输时滞和数据丢包转换为零阶保持器端数据更新周期, 在此基础上, 利用输入时滞法和PDC (Parallel distributed compensation) 技术, 建立网络环境下被控对象和参考模型合并的基于T-S 模糊模型的增广系统模型. 通过Lyapunov 方法, 并充分利用采样特性, 给出系统实现H输出跟踪的充分条件, 以及可靠模糊控制器的设计. 仿真结果表明所设计模糊控制器能够实现该类系统良好的跟踪.     

Variable precision intuitionistic fuzzy rough set model and applications based on conflict distance

Due to the complexity and uncertainty of the physical world, as well as the limitation of human ability to comprehend, it is very difficult for any single method of uncertainty to effectively deal with the decision‐making problem that exists in real life. So, it is natural for us to think about incorporating the advantages of various theories of uncertainty to develop a more powerful hybrid method of soft decision‐making. In view of this recognition, the thought and method of intuitionistic fuzzy sets and variable precision rough sets are used to construct a novel intuitionistic fuzzy rough set model. With respect to the fact that the information system is intuitionistic fuzzy, the idea of measuring intuitionistic fuzzy similarity is used to define conflict distance. After that, this concept is combined with the variable precision rough sets so that a variable precision intuitionistic fuzzy rough set model is established, and its properties are investigated. After proposing an attribute reduction algorithm based on variable precision intuitionistic fuzzy rough sets, a case study is used to verify the feasibility and effectiveness of our novel model. The results show that our model indeed improves the classification ability of earlier models and possesses some ability to tolerate faults through adjusting the parameter λ and the confidence threshold β; it realizes the correct classification and extracts the decision rules.