Harmony Search(HS)algorithm is highly effective in solving a wide range of real-world engineering optimization problems.However,it still has the problems such as being prone to local optima,low optimization accuracy,a...Harmony Search(HS)algorithm is highly effective in solving a wide range of real-world engineering optimization problems.However,it still has the problems such as being prone to local optima,low optimization accuracy,and low search efficiency.To address the limitations of the HS algorithm,a novel approach called the Dual-Memory Dynamic Search Harmony Search(DMDS-HS)algorithm is introduced.The main innovations of this algorithm are as follows:Firstly,a dual-memory structure is introduced to rank and hierarchically organize the harmonies in the harmony memory,creating an effective and selectable trust region to reduce approach blind searching.Furthermore,the trust region is dynamically adjusted to improve the convergence of the algorithm while maintaining its global search capability.Secondly,to boost the algorithm’s convergence speed,a phased dynamic convergence domain concept is introduced to strategically devise a global random search strategy.Lastly,the algorithm constructs an adaptive parameter adjustment strategy to adjust the usage probability of the algorithm’s search strategies,which aim to rationalize the abilities of exploration and exploitation of the algorithm.The results tested on the Computational Experiment Competition on 2017(CEC2017)test function set show that DMDS-HS outperforms the other nine HS algorithms and the other four state-of-the-art algorithms in terms of diversity,freedom from local optima,and solution accuracy.In addition,applying DMDS-HS to data clustering problems,the results show that it exhibits clustering performance that exceeds the other seven classical clustering algorithms,which verifies the effectiveness and reliability of DMDS-HS in solving complex data clustering problems.展开更多
现有的有监督可见光-近红外行人重识别方法需要大量人力资源去除手工标注数据,容易受到标注数据场景的限制,难以满足真实多变应用场景的泛化性。因此,文中提出基于语义伪标签和双重特征存储库的无监督跨模态行人重识别方法。首先,提出...现有的有监督可见光-近红外行人重识别方法需要大量人力资源去除手工标注数据,容易受到标注数据场景的限制,难以满足真实多变应用场景的泛化性。因此,文中提出基于语义伪标签和双重特征存储库的无监督跨模态行人重识别方法。首先,提出基于对比学习框架的预训练方法,利用可见光行人图像和其生成的辅助灰度图像进行训练。利用该预训练方法获取对颜色变化具有鲁棒性的语义特征提取网络。然后,使用DBSCAN(Density-Based Spatial Clustering of Applications with Noise)聚类方法生成语义伪标签。相比现有的伪标签生成方法,文中提出的语义伪标签在生成过程中充分利用跨模态数据之间的结构信息,减少跨模态数据颜色变化带来的模态差异。此外,文中还构建实例级困难样本特征存储库和中心级聚类特征存储库,充分利用困难样本特征和聚类特征,让模型对噪声伪标签具有更强的鲁棒性。在SYSU-MM01、RegDB两个跨模态数据集上的实验验证文中方法的有效性。展开更多
基金This work was supported by the Fund of Innovative Training Program for College Students of Guangzhou University(No.s202211078116)Guangzhou City School Joint Fund Project(No.SL2022A03J01009)+2 种基金National Natural Science Foundation of China(No.61806058)Natural Science Foundation of Guangdong Province(No.2018A030310063)Guangzhou Science and Technology Plan Project(No.201804010299).
文摘Harmony Search(HS)algorithm is highly effective in solving a wide range of real-world engineering optimization problems.However,it still has the problems such as being prone to local optima,low optimization accuracy,and low search efficiency.To address the limitations of the HS algorithm,a novel approach called the Dual-Memory Dynamic Search Harmony Search(DMDS-HS)algorithm is introduced.The main innovations of this algorithm are as follows:Firstly,a dual-memory structure is introduced to rank and hierarchically organize the harmonies in the harmony memory,creating an effective and selectable trust region to reduce approach blind searching.Furthermore,the trust region is dynamically adjusted to improve the convergence of the algorithm while maintaining its global search capability.Secondly,to boost the algorithm’s convergence speed,a phased dynamic convergence domain concept is introduced to strategically devise a global random search strategy.Lastly,the algorithm constructs an adaptive parameter adjustment strategy to adjust the usage probability of the algorithm’s search strategies,which aim to rationalize the abilities of exploration and exploitation of the algorithm.The results tested on the Computational Experiment Competition on 2017(CEC2017)test function set show that DMDS-HS outperforms the other nine HS algorithms and the other four state-of-the-art algorithms in terms of diversity,freedom from local optima,and solution accuracy.In addition,applying DMDS-HS to data clustering problems,the results show that it exhibits clustering performance that exceeds the other seven classical clustering algorithms,which verifies the effectiveness and reliability of DMDS-HS in solving complex data clustering problems.
文摘现有的有监督可见光-近红外行人重识别方法需要大量人力资源去除手工标注数据,容易受到标注数据场景的限制,难以满足真实多变应用场景的泛化性。因此,文中提出基于语义伪标签和双重特征存储库的无监督跨模态行人重识别方法。首先,提出基于对比学习框架的预训练方法,利用可见光行人图像和其生成的辅助灰度图像进行训练。利用该预训练方法获取对颜色变化具有鲁棒性的语义特征提取网络。然后,使用DBSCAN(Density-Based Spatial Clustering of Applications with Noise)聚类方法生成语义伪标签。相比现有的伪标签生成方法,文中提出的语义伪标签在生成过程中充分利用跨模态数据之间的结构信息,减少跨模态数据颜色变化带来的模态差异。此外,文中还构建实例级困难样本特征存储库和中心级聚类特征存储库,充分利用困难样本特征和聚类特征,让模型对噪声伪标签具有更强的鲁棒性。在SYSU-MM01、RegDB两个跨模态数据集上的实验验证文中方法的有效性。
