目的基于图像的驾驶员分心行为识别可认为是一种二级图像子分类问题,与传统的图像分类不同,驾驶员分心识别任务中的各类区别比较微小,如区分一幅图像是在弄头发还是打电话完全取决于驾驶员手上是否有手机这个物体,即图像中的较小区域就...目的基于图像的驾驶员分心行为识别可认为是一种二级图像子分类问题,与传统的图像分类不同,驾驶员分心识别任务中的各类区别比较微小,如区分一幅图像是在弄头发还是打电话完全取决于驾驶员手上是否有手机这个物体,即图像中的较小区域就决定了该图像的类别。对于那些图像差异较小的类别,通常的图像分类方法无法高精度地区分。因此,为了能够学习到不同驾驶行为之间微小的表征区别,提出了一种姿态引导的实例感知学习网络用于驾驶员行为识别。方法首先利用人体检测器检测到人体框,利用人体姿态估计获取具有辨识性的手部相关区域,将人体和手部区域的特征作为实例级别的特征,以此设计一种实例感知学习模块充分获取不同层级的上下文语义信息。其次利用手部相关特征构建双通道交互模块来对关键空间信息进行表征的同时,对视觉特征进行优化,组建成一个多分支的深度神经网络。最后将不同分支的结果进行融合。结果实验结果表明,本文方法在AUC(American University in Cairo)数据集和自建三客一危数据集上的测试准确率分别达到96.17%和96.97%,相较于未使用实例感知模块和通道交互的模型,准确率显著改善,在复杂数据集下识别效果提升明显。结论本文提出的姿态引导的实例感知学习网络,在一定程度上降低了环境的干扰,准确度高,能辅助驾驶员安全行车,减少交通事故的发生。展开更多
Objective In tongue diagnosis,the location,color,and distribution of spots can be used to speculate on the viscera and severity of the heat evil.This work focuses on the image analysis method of artificial intelligenc...Objective In tongue diagnosis,the location,color,and distribution of spots can be used to speculate on the viscera and severity of the heat evil.This work focuses on the image analysis method of artificial intelligence(AI)to study the spotted tongue recognition of traditional Chinese medicine(TCM).Methods A model of spotted tongue recognition and extraction is designed,which is based on the principle of image deep learning and instance segmentation.This model includes multiscale feature map generation,region proposal searching,and target region recognition.Firstly,deep convolution network is used to build multiscale low-and high-abstraction feature maps after which,target candidate box generation algorithm and selection strategy are used to select high-quality target candidate regions.Finally,classification network is used for classifying target regions and calculating target region pixels.As a result,the region segmentation of spotted tongue is obtained.Under non-standard illumination conditions,various tongue images were taken by mobile phones,and experiments were conducted.Results The spotted tongue recognition achieved an area under curve(AUC)of 92.40%,an accuracy of 84.30%with a sensitivity of 88.20%,a specificity of 94.19%,a recall of 88.20%,a regional pixel accuracy index pixel accuracy(PA)of 73.00%,a mean pixel accuracy(m PA)of73.00%,an intersection over union(Io U)of 60.00%,and a mean intersection over union(mIo U)of 56.00%.Conclusion The results of the study verify that the model is suitable for the application of the TCM tongue diagnosis system.Spotted tongue recognition via multiscale convolutional neural network(CNN)would help to improve spot classification and the accurate extraction of pixels of spot area as well as provide a practical method for intelligent tongue diagnosis of TCM.展开更多
The feature selection in analogy-based software effort estimation （ASEE） is formulized as a multi-objective optimization problem. One objective is designed to maximize the effort estimation accuracy and the other ob...The feature selection in analogy-based software effort estimation （ASEE） is formulized as a multi-objective optimization problem. One objective is designed to maximize the effort estimation accuracy and the other objective is designed to minimize the number of selected features. Based on these two potential conflict objectives, a novel wrapper- based feature selection method, multi-objective feature selection for analogy-based software effort estimation （MASE）, is proposed. In the empirical studies, 77 projects in Desharnais and 62 projects in Maxwell from the real world are selected as the evaluation objects and the proposed method MASE is compared with some baseline methods. Final results show that the proposed method can achieve better performance by selecting fewer features when considering MMRE （mean magnitude of relative error）, MdMRE （median magnitude of relative error）, PRED （ 0. 25 ）, and SA （ standardized accuracy） performance metrics.展开更多
Complementing our previous publications, this paper presents the information schema constructs （ISCs） that underpin the programming of specific system manifestation feature （SMF） orientated information management ...Complementing our previous publications, this paper presents the information schema constructs （ISCs） that underpin the programming of specific system manifestation feature （SMF） orientated information management and composing system models. First, we briefly present （1） the general process of pre-embodiment design with SMFs, （2） the procedures of creating genotypes and phenotypes of SMFs, （3） the specific procedure of instantiation of phenotypes of SMFs, and （4） the procedure of system model management and processing. Then, the chunks of information needed for instantiation of phenotypes of SMFs are discussed, and the ISCs designed for instantiation presented. Afterwards, the information management aspects of system modeling are addressed. Methodologically, system modeling involves （1） placement of phenotypes of SMF in the modeling space, （2） combining them towards the desired architecture and operation, （3） assigning values to the parameters and checking the satisfac- tion of constraints, and （4） storing the system model in the SMFs-based warehouse database. The final objective of the reported research is to develop an SMFs-based toolbox to support modeling of cyber-physical systems （CPSs）.展开更多
文摘目的基于图像的驾驶员分心行为识别可认为是一种二级图像子分类问题,与传统的图像分类不同,驾驶员分心识别任务中的各类区别比较微小,如区分一幅图像是在弄头发还是打电话完全取决于驾驶员手上是否有手机这个物体,即图像中的较小区域就决定了该图像的类别。对于那些图像差异较小的类别,通常的图像分类方法无法高精度地区分。因此,为了能够学习到不同驾驶行为之间微小的表征区别,提出了一种姿态引导的实例感知学习网络用于驾驶员行为识别。方法首先利用人体检测器检测到人体框,利用人体姿态估计获取具有辨识性的手部相关区域,将人体和手部区域的特征作为实例级别的特征,以此设计一种实例感知学习模块充分获取不同层级的上下文语义信息。其次利用手部相关特征构建双通道交互模块来对关键空间信息进行表征的同时,对视觉特征进行优化,组建成一个多分支的深度神经网络。最后将不同分支的结果进行融合。结果实验结果表明,本文方法在AUC(American University in Cairo)数据集和自建三客一危数据集上的测试准确率分别达到96.17%和96.97%,相较于未使用实例感知模块和通道交互的模型,准确率显著改善,在复杂数据集下识别效果提升明显。结论本文提出的姿态引导的实例感知学习网络,在一定程度上降低了环境的干扰,准确度高,能辅助驾驶员安全行车,减少交通事故的发生。
基金Anhui Province College Natural Science Fund Key Project of China(KJ2020ZD77)the Project of Education Department of Anhui Province(KJ2020A0379)。
文摘Objective In tongue diagnosis,the location,color,and distribution of spots can be used to speculate on the viscera and severity of the heat evil.This work focuses on the image analysis method of artificial intelligence(AI)to study the spotted tongue recognition of traditional Chinese medicine(TCM).Methods A model of spotted tongue recognition and extraction is designed,which is based on the principle of image deep learning and instance segmentation.This model includes multiscale feature map generation,region proposal searching,and target region recognition.Firstly,deep convolution network is used to build multiscale low-and high-abstraction feature maps after which,target candidate box generation algorithm and selection strategy are used to select high-quality target candidate regions.Finally,classification network is used for classifying target regions and calculating target region pixels.As a result,the region segmentation of spotted tongue is obtained.Under non-standard illumination conditions,various tongue images were taken by mobile phones,and experiments were conducted.Results The spotted tongue recognition achieved an area under curve(AUC)of 92.40%,an accuracy of 84.30%with a sensitivity of 88.20%,a specificity of 94.19%,a recall of 88.20%,a regional pixel accuracy index pixel accuracy(PA)of 73.00%,a mean pixel accuracy(m PA)of73.00%,an intersection over union(Io U)of 60.00%,and a mean intersection over union(mIo U)of 56.00%.Conclusion The results of the study verify that the model is suitable for the application of the TCM tongue diagnosis system.Spotted tongue recognition via multiscale convolutional neural network(CNN)would help to improve spot classification and the accurate extraction of pixels of spot area as well as provide a practical method for intelligent tongue diagnosis of TCM.
基金The National Natural Science Foundation of China(No.61602267,61202006)the Open Project of State Key Laboratory for Novel Software Technology at Nanjing University(No.KFKT2016B18)
文摘The feature selection in analogy-based software effort estimation （ASEE） is formulized as a multi-objective optimization problem. One objective is designed to maximize the effort estimation accuracy and the other objective is designed to minimize the number of selected features. Based on these two potential conflict objectives, a novel wrapper- based feature selection method, multi-objective feature selection for analogy-based software effort estimation （MASE）, is proposed. In the empirical studies, 77 projects in Desharnais and 62 projects in Maxwell from the real world are selected as the evaluation objects and the proposed method MASE is compared with some baseline methods. Final results show that the proposed method can achieve better performance by selecting fewer features when considering MMRE （mean magnitude of relative error）, MdMRE （median magnitude of relative error）, PRED （ 0. 25 ）, and SA （ standardized accuracy） performance metrics.
文摘Complementing our previous publications, this paper presents the information schema constructs （ISCs） that underpin the programming of specific system manifestation feature （SMF） orientated information management and composing system models. First, we briefly present （1） the general process of pre-embodiment design with SMFs, （2） the procedures of creating genotypes and phenotypes of SMFs, （3） the specific procedure of instantiation of phenotypes of SMFs, and （4） the procedure of system model management and processing. Then, the chunks of information needed for instantiation of phenotypes of SMFs are discussed, and the ISCs designed for instantiation presented. Afterwards, the information management aspects of system modeling are addressed. Methodologically, system modeling involves （1） placement of phenotypes of SMF in the modeling space, （2） combining them towards the desired architecture and operation, （3） assigning values to the parameters and checking the satisfac- tion of constraints, and （4） storing the system model in the SMFs-based warehouse database. The final objective of the reported research is to develop an SMFs-based toolbox to support modeling of cyber-physical systems （CPSs）.
