This article studies the cooperative search-attack mission problem with dynamic targets and threats, and presents a Distributed Intelligent Self-Organized Mission Planning(DISOMP)algorithm for multiple Unmanned Aerial...This article studies the cooperative search-attack mission problem with dynamic targets and threats, and presents a Distributed Intelligent Self-Organized Mission Planning(DISOMP)algorithm for multiple Unmanned Aerial Vehicles(multi-UAV). The DISOMP algorithm can be divided into four modules: a search module designed based on the distributed Ant Colony Optimization(ACO) algorithm, an attack module designed based on the Parallel Approach(PA)scheme, a threat avoidance module designed based on the Dubins Curve(DC) and a communication module designed for information exchange among the multi-UAV system and the dynamic environment. A series of simulations of multi-UAV searching and attacking the moving targets are carried out, in which the search-attack mission completeness, execution efficiency and system suitability of the DISOMP algorithm are analyzed. The simulation results exhibit that the DISOMP algorithm based on online distributed down-top strategy is characterized by good flexibility, scalability and adaptability, in the dynamic targets searching and attacking problem.展开更多
Cooperative search-attack is an important application of unmanned aerial vehicle(UAV)swarm in military field.The coupling between path planning and task allocation,the heterogeneity of UAVs,and the dynamic nature of t...Cooperative search-attack is an important application of unmanned aerial vehicle(UAV)swarm in military field.The coupling between path planning and task allocation,the heterogeneity of UAVs,and the dynamic nature of task environment greatly increase the complexity and difficulty of the UAV swarm cooperative search-attack mission planning problem.Inspired by the collaborative hunting behavior of wolf pack,a distributed selforganizing method for UAV swarm search-attack mission planning is proposed.First,to solve the multi-target search problem in unknown environments,a wolf scouting behavior-inspired cooperative search algorithm for UAV swarm is designed.Second,a distributed self-organizing task allocation algorithm for UAV swarm cooperative attacking of targets is proposed by analyzing the flexible labor division behavior of wolves.By abstracting the UAV as a simple artificial wolf agent,the flexible motion planning and group task coordinating for UAV swarm can be realized by self-organizing.The effectiveness of the proposed method is verified by a set of simulation experiments,the stability and scalability are evaluated,and the integrated solution for the coupled path planning and task allocation problems for the UAV swarm cooperative search-attack task can be well performed.展开更多
深度神经网络(DNN)极易受到对抗样本的影响,仅需向原始文本中添加细微的扰动即可诱导目标模型做出误判。研究对抗样本的生成不仅有利于提升模型的鲁棒性,还能推动DNN可解释性方面的工作。在中文对抗领域,现有的中文对抗样本生成方法大...深度神经网络(DNN)极易受到对抗样本的影响,仅需向原始文本中添加细微的扰动即可诱导目标模型做出误判。研究对抗样本的生成不仅有利于提升模型的鲁棒性,还能推动DNN可解释性方面的工作。在中文对抗领域,现有的中文对抗样本生成方法大多采用单一变换策略,仅考虑了部分汉语特征,并且忽视了攻击对上下文语境产生的影响。为了解决这些问题,提出一种基于启发式算法的中文对抗样本生成方法BSCA。通过全面分析表音文字和意音文字之间的差异,结合汉语的构字法、字音、字形、认知语言学等先验知识,设计可准确评估汉字差异的中文文本扰动策略。利用扰动策略构建对抗搜索空间,并运用改进的集束搜索算法对黑盒攻击过程进行优化。在严格限制扰动大小和语义偏移的情况下,BSCA能够自动选择不同的攻击策略,以适应不同场景需求。在多个自然语言处理(NLP)任务上分别对TextCNN、TextRNN和BERT(Bidirectional Encoder Representations from Transformers)模型进行实验,结果表明,BSCA具有较好的泛化能力,能使分类准确率至少降低63.84百分点,同时拥有比基线方法更低的攻击代价。展开更多
基金supported in part by National Natural Science Foundation of China (Nos. 61741313, 61673209, and 61533008)Jiangsu Six Peak of Talents Program, China (No. KTHY-027)Postgraduate Research & Practice Innovation Program of Jiangsu Province, China (No. KYCX18_0303)
文摘This article studies the cooperative search-attack mission problem with dynamic targets and threats, and presents a Distributed Intelligent Self-Organized Mission Planning(DISOMP)algorithm for multiple Unmanned Aerial Vehicles(multi-UAV). The DISOMP algorithm can be divided into four modules: a search module designed based on the distributed Ant Colony Optimization(ACO) algorithm, an attack module designed based on the Parallel Approach(PA)scheme, a threat avoidance module designed based on the Dubins Curve(DC) and a communication module designed for information exchange among the multi-UAV system and the dynamic environment. A series of simulations of multi-UAV searching and attacking the moving targets are carried out, in which the search-attack mission completeness, execution efficiency and system suitability of the DISOMP algorithm are analyzed. The simulation results exhibit that the DISOMP algorithm based on online distributed down-top strategy is characterized by good flexibility, scalability and adaptability, in the dynamic targets searching and attacking problem.
基金supported by the National Natural Science Foundation of China(61502534)the Shaanxi Provincial Natural Science Foundation(2020JQ-493)+2 种基金the Integrative Equipment Research Project of Armed Police Force(WJ20211A030018)the Military Science Project of the National Social Science Fund(WJ2019-SKJJ-C-092)the Theoretical Research Foundation of Armed Police Engineering University(WJY202148)。
文摘Cooperative search-attack is an important application of unmanned aerial vehicle(UAV)swarm in military field.The coupling between path planning and task allocation,the heterogeneity of UAVs,and the dynamic nature of task environment greatly increase the complexity and difficulty of the UAV swarm cooperative search-attack mission planning problem.Inspired by the collaborative hunting behavior of wolf pack,a distributed selforganizing method for UAV swarm search-attack mission planning is proposed.First,to solve the multi-target search problem in unknown environments,a wolf scouting behavior-inspired cooperative search algorithm for UAV swarm is designed.Second,a distributed self-organizing task allocation algorithm for UAV swarm cooperative attacking of targets is proposed by analyzing the flexible labor division behavior of wolves.By abstracting the UAV as a simple artificial wolf agent,the flexible motion planning and group task coordinating for UAV swarm can be realized by self-organizing.The effectiveness of the proposed method is verified by a set of simulation experiments,the stability and scalability are evaluated,and the integrated solution for the coupled path planning and task allocation problems for the UAV swarm cooperative search-attack task can be well performed.
文摘深度神经网络(DNN)极易受到对抗样本的影响,仅需向原始文本中添加细微的扰动即可诱导目标模型做出误判。研究对抗样本的生成不仅有利于提升模型的鲁棒性,还能推动DNN可解释性方面的工作。在中文对抗领域,现有的中文对抗样本生成方法大多采用单一变换策略,仅考虑了部分汉语特征,并且忽视了攻击对上下文语境产生的影响。为了解决这些问题,提出一种基于启发式算法的中文对抗样本生成方法BSCA。通过全面分析表音文字和意音文字之间的差异,结合汉语的构字法、字音、字形、认知语言学等先验知识,设计可准确评估汉字差异的中文文本扰动策略。利用扰动策略构建对抗搜索空间,并运用改进的集束搜索算法对黑盒攻击过程进行优化。在严格限制扰动大小和语义偏移的情况下,BSCA能够自动选择不同的攻击策略,以适应不同场景需求。在多个自然语言处理(NLP)任务上分别对TextCNN、TextRNN和BERT(Bidirectional Encoder Representations from Transformers)模型进行实验,结果表明,BSCA具有较好的泛化能力,能使分类准确率至少降低63.84百分点,同时拥有比基线方法更低的攻击代价。
