目的当前,基于视觉的步态识别方法多基于完整的步态序列图像。然而,现实场景拍摄下的行人难免被遮挡,以至于获取的步态图像不完整,对识别结果有很大影响。如何处理大面积遮挡是步态识别中一个具有挑战性且重要的问题。针对此,提出了一...目的当前,基于视觉的步态识别方法多基于完整的步态序列图像。然而,现实场景拍摄下的行人难免被遮挡,以至于获取的步态图像不完整,对识别结果有很大影响。如何处理大面积遮挡是步态识别中一个具有挑战性且重要的问题。针对此,提出了一种步态时空序列重建网络(gait spatio-temporal reconstruction network,GSTRNet),用于修复被遮挡的步态序列图像。方法使用基于3D卷积神经网络和Transformer的GSTRNet来修复步态序列,在修复每一帧步态图像的空间信息的同时保持帧与帧之间的时空连贯性。GSTRNet通过引入YOLOv5(you only look once)网络来检测步态图像的局部遮挡区域,并将其作为先验知识为遮挡修复区域分配更高的修复权值,实现遮挡区域的局部修复,将局部修复步态图与原始遮挡图像进行融合,生成完整的修复步态图。同时,在GSTRNet中引入三元组特征损失和重建损失组成的联合损失函数来优化修复网络,提升修复效果。最终,以修复完整的步态序列图像为特征进行身份识别。结果本文在大规模步态数据集OU_MVLP(the OU-ISIR gait database,multi-view large population dataset)中人工合成遮挡步态序列数据来进行修复实验。结果表明,该方法在面对步态轮廓大面积遮挡时,识别准确率比现有的步态修复和遮挡识别方法有一定的提升,如在未知遮挡模式时比三元组视频生成对抗网络(sequence video wasserstein generative adversarial network based on triplet hinge loss,sVideoWGAN-hinge)最高提升6.7%,非单一模式遮挡时比Gaitset等方法识别率提高40%左右。结论本文提出的GSTRNet对各种遮挡模式下的步态图像序列有较好的修复效果,使用修复后图像进行步态识别,可有效改善识别率。展开更多
Purpose-Aiming at the shortcomings of EEG signals generated by brain’s sensorimotor region activated tasks,such as poor performance,low efficiency and weak robustness,this paper proposes an EEG signals classification...Purpose-Aiming at the shortcomings of EEG signals generated by brain’s sensorimotor region activated tasks,such as poor performance,low efficiency and weak robustness,this paper proposes an EEG signals classification method based on multi-dimensional fusion features.Design/methodology/approach-First,the improved Morlet wavelet is used to extract the spectrum feature maps from EEG signals.Then,the spatial-frequency features are extracted from the PSD maps by using the three-dimensional convolutional neural networks(3DCNNs)model.Finally,the spatial-frequency features are incorporated to the bidirectional gated recurrent units(Bi-GRUs)models to extract the spatial-frequencysequential multi-dimensional fusion features for recognition of brain’s sensorimotor region activated task.Findings-In the comparative experiments,the data sets of motor imagery(MI)/action observation(AO)/action execution(AE)tasks are selected to test the classification performance and robustness of the proposed algorithm.In addition,the impact of extracted features on the sensorimotor region and the impact on the classification processing are also analyzed by visualization during experiments.Originality/value-The experimental results show that the proposed algorithm extracts the corresponding brain activation features for different action related tasks,so as to achieve more stable classification performance in dealing with AO/MI/AE tasks,and has the best robustness on EEGsignals of different subjects.展开更多
文摘【目的】预测土地利用变化可以为在资源分配、环境保护、城镇化发展、灾害防范与风险评估等方面提供合理依据,有利于制定科学的政策。现有的元胞自动机(Cellular Automata,CA)模型在预测土地利用变化中,对多时序复杂土地利用变化中时空特征提取不足,难以适用于复杂土地利用变化的精准预测。【方法】本文提出了一种融合注意力机制并考虑多尺度驱动因素的深度时空建模网络(3DCBLT),并与CA模型耦合,旨在优化时空特征提取与非线性建模能力。3DCBLT整合了跨通道、空间注意力机制(Convolutional Block Attention Module,CBAM)和三维卷积(3D Convolution,3DCNN)模块旨在强化对土地利用变化关键区域的关注,并提升对时空特征的深层提取能力。同时,采用长短时记忆网络(Long Short Term Memory Network,LSTM)充分挖掘土地利用演化过程中的时间依赖性与长期趋势,得到不同地类发展概率。本文以陕西省作为研究区,选取2000—2020年每5年一期土地利用数据并引入气候、地形、经济等12项指标作为驱动因素,以2020年土地利用数据为验证。【结果】本文提出模型的Kappa系数为0.888,OA系数为0.925,FoM系数为0.336,显著优于ANN-CA、MLP-CA和CAMarko,验证了该模型的在土地利用变化预测的有效性与优越性。【结论】本文提出的3DCBLT-CA模型在复杂土地利用变化预测中表现出较高的精度和优越的时空建模能力,为复杂土地利用情景下的变化模拟提供了一种切实可行的技术路径。
文摘目的当前,基于视觉的步态识别方法多基于完整的步态序列图像。然而,现实场景拍摄下的行人难免被遮挡,以至于获取的步态图像不完整,对识别结果有很大影响。如何处理大面积遮挡是步态识别中一个具有挑战性且重要的问题。针对此,提出了一种步态时空序列重建网络(gait spatio-temporal reconstruction network,GSTRNet),用于修复被遮挡的步态序列图像。方法使用基于3D卷积神经网络和Transformer的GSTRNet来修复步态序列,在修复每一帧步态图像的空间信息的同时保持帧与帧之间的时空连贯性。GSTRNet通过引入YOLOv5(you only look once)网络来检测步态图像的局部遮挡区域,并将其作为先验知识为遮挡修复区域分配更高的修复权值,实现遮挡区域的局部修复,将局部修复步态图与原始遮挡图像进行融合,生成完整的修复步态图。同时,在GSTRNet中引入三元组特征损失和重建损失组成的联合损失函数来优化修复网络,提升修复效果。最终,以修复完整的步态序列图像为特征进行身份识别。结果本文在大规模步态数据集OU_MVLP(the OU-ISIR gait database,multi-view large population dataset)中人工合成遮挡步态序列数据来进行修复实验。结果表明,该方法在面对步态轮廓大面积遮挡时,识别准确率比现有的步态修复和遮挡识别方法有一定的提升,如在未知遮挡模式时比三元组视频生成对抗网络(sequence video wasserstein generative adversarial network based on triplet hinge loss,sVideoWGAN-hinge)最高提升6.7%,非单一模式遮挡时比Gaitset等方法识别率提高40%左右。结论本文提出的GSTRNet对各种遮挡模式下的步态图像序列有较好的修复效果,使用修复后图像进行步态识别,可有效改善识别率。
基金The education and scientific research project of young and middle-aged teachers of Fujian provincial department of education(No.JAT171070).
文摘Purpose-Aiming at the shortcomings of EEG signals generated by brain’s sensorimotor region activated tasks,such as poor performance,low efficiency and weak robustness,this paper proposes an EEG signals classification method based on multi-dimensional fusion features.Design/methodology/approach-First,the improved Morlet wavelet is used to extract the spectrum feature maps from EEG signals.Then,the spatial-frequency features are extracted from the PSD maps by using the three-dimensional convolutional neural networks(3DCNNs)model.Finally,the spatial-frequency features are incorporated to the bidirectional gated recurrent units(Bi-GRUs)models to extract the spatial-frequencysequential multi-dimensional fusion features for recognition of brain’s sensorimotor region activated task.Findings-In the comparative experiments,the data sets of motor imagery(MI)/action observation(AO)/action execution(AE)tasks are selected to test the classification performance and robustness of the proposed algorithm.In addition,the impact of extracted features on the sensorimotor region and the impact on the classification processing are also analyzed by visualization during experiments.Originality/value-The experimental results show that the proposed algorithm extracts the corresponding brain activation features for different action related tasks,so as to achieve more stable classification performance in dealing with AO/MI/AE tasks,and has the best robustness on EEGsignals of different subjects.
