针对原始SSD(Single Shot Multibox Detector)算法未充分利用各特征层之间关系导致浅层特征层缺乏小目标语义信息的问题,为了提高对小目标的检测能力,提出了一种结合PANet多尺度特征融合网络和自上向下特征融合路径的TTB-SSD(Top to Bot...针对原始SSD(Single Shot Multibox Detector)算法未充分利用各特征层之间关系导致浅层特征层缺乏小目标语义信息的问题,为了提高对小目标的检测能力,提出了一种结合PANet多尺度特征融合网络和自上向下特征融合路径的TTB-SSD(Top to Bottom SSD)改进算法。首先,使用PANet多尺度特征融合网络对特征进行反复提取,从而获得丰富的多尺度语义信息;然后,使用一种深层特征融合模块将浅层特征层的空间信息传递到深层特征层,进而更准确地对小目标进行定位;最后,为了增强浅层特征层的语义信息,构造了自上向下的特征融合路径,从而强化浅层对小目标检测的准确率。实验结果表明,在PASCAL VOC2007测试集检测的mAP(Mean Average Precision)值达到80.5%,对目标的mAP较原始SSD提高了5.7%,证明了该算法对小目标检测的有效性。展开更多
Top‐down synthesis has been used to prepare catalytic materials with nanometer sizes,but fabricating atomically dispersed metal catalysts remains a challenge because surface single metal atoms are prone to aggregatio...Top‐down synthesis has been used to prepare catalytic materials with nanometer sizes,but fabricating atomically dispersed metal catalysts remains a challenge because surface single metal atoms are prone to aggregation or coalescence.A top‐down strategy is used to synthesize atomically dispersed metal catalysts,based on supported Ag nanoparticles.The changes of the geometric and electronic structures of the Ag atoms during the top‐down process are studied using the in situ synchrotron X‐ray diffraction technique,ex situ X‐ray absorption spectroscopy,and transmission electron microscopy.The experimental results,coupled with the density functional theory calculations,demonstrate that the electronic perturbation of the Ag frontier orbitals,induced by the Ag‐O interactions at the perimeter of the metal‐support interface,is the driving force of the top‐down process.The top‐down synthesis has two important functions:to increase the number of catalytic active sites and to facilitate the study of complex reaction mechanisms(e.g.,formaldehyde oxidation)by developing single‐site model catalysts.展开更多
文摘针对原始SSD(Single Shot Multibox Detector)算法未充分利用各特征层之间关系导致浅层特征层缺乏小目标语义信息的问题,为了提高对小目标的检测能力,提出了一种结合PANet多尺度特征融合网络和自上向下特征融合路径的TTB-SSD(Top to Bottom SSD)改进算法。首先,使用PANet多尺度特征融合网络对特征进行反复提取,从而获得丰富的多尺度语义信息;然后,使用一种深层特征融合模块将浅层特征层的空间信息传递到深层特征层,进而更准确地对小目标进行定位;最后,为了增强浅层特征层的语义信息,构造了自上向下的特征融合路径,从而强化浅层对小目标检测的准确率。实验结果表明,在PASCAL VOC2007测试集检测的mAP(Mean Average Precision)值达到80.5%,对目标的mAP较原始SSD提高了5.7%,证明了该算法对小目标检测的有效性。
基金supported by the National Natural Science Foundation of China(21477023)the Science and Technology Commission of Shanghai Municipality(14JC1400400)~~
文摘Top‐down synthesis has been used to prepare catalytic materials with nanometer sizes,but fabricating atomically dispersed metal catalysts remains a challenge because surface single metal atoms are prone to aggregation or coalescence.A top‐down strategy is used to synthesize atomically dispersed metal catalysts,based on supported Ag nanoparticles.The changes of the geometric and electronic structures of the Ag atoms during the top‐down process are studied using the in situ synchrotron X‐ray diffraction technique,ex situ X‐ray absorption spectroscopy,and transmission electron microscopy.The experimental results,coupled with the density functional theory calculations,demonstrate that the electronic perturbation of the Ag frontier orbitals,induced by the Ag‐O interactions at the perimeter of the metal‐support interface,is the driving force of the top‐down process.The top‐down synthesis has two important functions:to increase the number of catalytic active sites and to facilitate the study of complex reaction mechanisms(e.g.,formaldehyde oxidation)by developing single‐site model catalysts.
