人脸识别技术广泛应用于考勤管理、移动支付等智慧建设中。伴随着常态化的口罩干扰,传统人脸识别算法已无法满足实际应用需求,为此,本文利用深度学习模型SSD以及FaceNet模型对人脸识别系统展开设计。首先,为消除现有数据集中亚洲人脸占...人脸识别技术广泛应用于考勤管理、移动支付等智慧建设中。伴随着常态化的口罩干扰,传统人脸识别算法已无法满足实际应用需求,为此,本文利用深度学习模型SSD以及FaceNet模型对人脸识别系统展开设计。首先,为消除现有数据集中亚洲人脸占比小造成的类内间距变化差距不明显的问题,在CAS-IA Web Face公开数据集的基础上对亚洲人脸数据进行扩充;其次,为解决不同口罩样式对特征提取的干扰,使用SSD人脸检测模型与DLIB人脸关键点检测模型提取人脸关键点,并利用人脸关键点与口罩的空间位置关系,额外随机生成不同的口罩人脸,组成混合数据集;最后,在混合数据集上进行模型训练并将训练好的模型移植到人脸识别系统中,进行检测速度与识别精度验证。实验结果表明,系统的实时识别速度达20 fps以上,人脸识别模型准确率在构建的混合数据集中达到97.1%,在随机抽取的部分LFW数据集验证的准确率达99.7%,故而该系统可满足实际应用需求,在一定程度上提高人脸识别的鲁棒性与准确性。展开更多
Interest rates are the key to the resource allocations of financial markets.The frequent appearance of negative nominal interest rates(NNIR)may lead to a failure of the transmission mechanism and arouse system risks i...Interest rates are the key to the resource allocations of financial markets.The frequent appearance of negative nominal interest rates(NNIR)may lead to a failure of the transmission mechanism and arouse system risks in the financial markets.Meanwhile,negative nominal interest rates is a new policy with no consensus reached by academia or policymakers.It is necessary to review the research results and promote consensus to reveal the nature and impact mechanisms of NNIR.Therefore,we retrieved original articles from the Web of Science(WoS)and China National Knowledge Infrastructure(CNKI)published from 1999 to 2020 on NNIR to determine the characteristics of current research results from various perspectives and compare literature in English and Chinese of highly productive institutions and researchers,hot topics,evolution contexts,research frontiers,and ecological characteristics.There are three major findings.In terms of research ecology,the ecological structure of the top research institutions,both domestic and abroad,remains steady,but the structure of the researchers is not stable.In terms of a research domain,NNIR is studied from many dimensions,and the early established research domains still have long-lasting impacts in English literature.In contrast,Chinese literature mainly focuses on discussing applicable policies with no highly focused domain and research topic with great influence yet.Finally,the focus in both English and Chinese literature has changed.Since the original research framework is insufficient in explanatory power,literature in English is beginning to pay attention to the empirical analyses of practical policies,and Chinese scholars are turning to theoretical study to enhance the in-depth understanding of this phenomenon.Researchers in China should focus on collecting high-quality research materials and pay more attention to the progress of empirical research in English literature to improve Chinese research efficiency and quality and then promote research progress in China.展开更多
文摘人脸识别技术广泛应用于考勤管理、移动支付等智慧建设中。伴随着常态化的口罩干扰,传统人脸识别算法已无法满足实际应用需求,为此,本文利用深度学习模型SSD以及FaceNet模型对人脸识别系统展开设计。首先,为消除现有数据集中亚洲人脸占比小造成的类内间距变化差距不明显的问题,在CAS-IA Web Face公开数据集的基础上对亚洲人脸数据进行扩充;其次,为解决不同口罩样式对特征提取的干扰,使用SSD人脸检测模型与DLIB人脸关键点检测模型提取人脸关键点,并利用人脸关键点与口罩的空间位置关系,额外随机生成不同的口罩人脸,组成混合数据集;最后,在混合数据集上进行模型训练并将训练好的模型移植到人脸识别系统中,进行检测速度与识别精度验证。实验结果表明,系统的实时识别速度达20 fps以上,人脸识别模型准确率在构建的混合数据集中达到97.1%,在随机抽取的部分LFW数据集验证的准确率达99.7%,故而该系统可满足实际应用需求,在一定程度上提高人脸识别的鲁棒性与准确性。
基金Planning Projects on Philosophy and Social Sciences of Shanghai(2020BJB020).
文摘Interest rates are the key to the resource allocations of financial markets.The frequent appearance of negative nominal interest rates(NNIR)may lead to a failure of the transmission mechanism and arouse system risks in the financial markets.Meanwhile,negative nominal interest rates is a new policy with no consensus reached by academia or policymakers.It is necessary to review the research results and promote consensus to reveal the nature and impact mechanisms of NNIR.Therefore,we retrieved original articles from the Web of Science(WoS)and China National Knowledge Infrastructure(CNKI)published from 1999 to 2020 on NNIR to determine the characteristics of current research results from various perspectives and compare literature in English and Chinese of highly productive institutions and researchers,hot topics,evolution contexts,research frontiers,and ecological characteristics.There are three major findings.In terms of research ecology,the ecological structure of the top research institutions,both domestic and abroad,remains steady,but the structure of the researchers is not stable.In terms of a research domain,NNIR is studied from many dimensions,and the early established research domains still have long-lasting impacts in English literature.In contrast,Chinese literature mainly focuses on discussing applicable policies with no highly focused domain and research topic with great influence yet.Finally,the focus in both English and Chinese literature has changed.Since the original research framework is insufficient in explanatory power,literature in English is beginning to pay attention to the empirical analyses of practical policies,and Chinese scholars are turning to theoretical study to enhance the in-depth understanding of this phenomenon.Researchers in China should focus on collecting high-quality research materials and pay more attention to the progress of empirical research in English literature to improve Chinese research efficiency and quality and then promote research progress in China.
