针对通用目标检测场景下,现有单阶段无锚检测器识别精度低、识别困难等问题,提出一种基于改进变焦网络VFNet(VarifocalNet)的高精度目标检测算法。首先,利用循环层聚合网络(RLANet)替换VFNet用于特征提取的主干网络ResNet,循环残差连接...针对通用目标检测场景下,现有单阶段无锚检测器识别精度低、识别困难等问题,提出一种基于改进变焦网络VFNet(VarifocalNet)的高精度目标检测算法。首先,利用循环层聚合网络(RLANet)替换VFNet用于特征提取的主干网络ResNet,循环残差连接操作将前层特征汇入后续网络层中提升特征的表征能力;其次,通过带有特征对齐卷积操作的特征金字塔网络(FPN)替换原始的特征融合网络,利用可变形卷积操作在FPN上下层融合过程中实现特征对齐并优化特征表征能力;最后,使用聚焦-全局蒸馏(FGD)算法进一步提升小规模算法的检测性能。在COCO(Common Objects in Context)2017数据集上进行的评估实验结果表明,在相同训练条件下,改进后的以RLANet-50为主干的算法的均值平均精度(mAP)可以达到45.9%,与VFNet算法相比提升了4.3个百分点,而改进后的算法参数量为36.67×10^(6),与VFNet相比仅高了4×10^(6)。可见,改进后的VFNet算法在提升检测精度的同时稍微增加了参数量,说明该算法可以满足目标检测的轻量化及高精度需求。展开更多
The electro-polymerization behavior of aniline in reverse(W/O) microemulsion was investigated. The experiment results show that the cyclic voltammetry polymerization behavior of aniline in W/O microemulsion is differe...The electro-polymerization behavior of aniline in reverse(W/O) microemulsion was investigated. The experiment results show that the cyclic voltammetry polymerization behavior of aniline in W/O microemulsion is different from that in aqueous solution remarkably. With the increase of scan cycle, the oxidation potential shifts positively and the reduction potential shifts negatively, i.e., the redox potential difference increases. H+ apparent concentration affects the aniline polymerization evidently. When H+ concentration is lower than 0.08 mol/L, the electro-polymerization of aniline is difficult. With the increase of H+ concentration, the polymerization current of aniline increases gradually. Only when H+ concentration is high enough(0.5 mol/L), aniline can be well electro-polymerized. Moreover, under the same condition, the aniline polymerization current in W/O microemulsion is higher than that in aqueous solution. The scanning electron microscopy image shows that the deposited polyaniline(PANI) has uniform fiber morphology with diameter of about 100 nm. Further study result suggests that the electrochemical activity of the PANI in HCl is similar to that of the PANI prepared in aqueous solution.展开更多
文摘针对通用目标检测场景下,现有单阶段无锚检测器识别精度低、识别困难等问题,提出一种基于改进变焦网络VFNet(VarifocalNet)的高精度目标检测算法。首先,利用循环层聚合网络(RLANet)替换VFNet用于特征提取的主干网络ResNet,循环残差连接操作将前层特征汇入后续网络层中提升特征的表征能力;其次,通过带有特征对齐卷积操作的特征金字塔网络(FPN)替换原始的特征融合网络,利用可变形卷积操作在FPN上下层融合过程中实现特征对齐并优化特征表征能力;最后,使用聚焦-全局蒸馏(FGD)算法进一步提升小规模算法的检测性能。在COCO(Common Objects in Context)2017数据集上进行的评估实验结果表明,在相同训练条件下,改进后的以RLANet-50为主干的算法的均值平均精度(mAP)可以达到45.9%,与VFNet算法相比提升了4.3个百分点,而改进后的算法参数量为36.67×10^(6),与VFNet相比仅高了4×10^(6)。可见,改进后的VFNet算法在提升检测精度的同时稍微增加了参数量,说明该算法可以满足目标检测的轻量化及高精度需求。
基金Projects(51071067,21271069,20673036,J1210040,50473022) supported by National Natural Science Foundation of ChinaProject(2013GK3015) supported by the Science and Technology Program of Hunan Province,China
文摘The electro-polymerization behavior of aniline in reverse(W/O) microemulsion was investigated. The experiment results show that the cyclic voltammetry polymerization behavior of aniline in W/O microemulsion is different from that in aqueous solution remarkably. With the increase of scan cycle, the oxidation potential shifts positively and the reduction potential shifts negatively, i.e., the redox potential difference increases. H+ apparent concentration affects the aniline polymerization evidently. When H+ concentration is lower than 0.08 mol/L, the electro-polymerization of aniline is difficult. With the increase of H+ concentration, the polymerization current of aniline increases gradually. Only when H+ concentration is high enough(0.5 mol/L), aniline can be well electro-polymerized. Moreover, under the same condition, the aniline polymerization current in W/O microemulsion is higher than that in aqueous solution. The scanning electron microscopy image shows that the deposited polyaniline(PANI) has uniform fiber morphology with diameter of about 100 nm. Further study result suggests that the electrochemical activity of the PANI in HCl is similar to that of the PANI prepared in aqueous solution.
