针对航拍图像易受雾气影响,AOD-Net(All in one dehazing network)算法对图像去雾后容易出现细节模糊、对比度过高和图像偏暗等问题,本文提出了一种基于改进AOD-Net的航拍图像去雾算法.本文主要从网络结构、损失函数、训练方式三个方面...针对航拍图像易受雾气影响,AOD-Net(All in one dehazing network)算法对图像去雾后容易出现细节模糊、对比度过高和图像偏暗等问题,本文提出了一种基于改进AOD-Net的航拍图像去雾算法.本文主要从网络结构、损失函数、训练方式三个方面对AOD-Net进行改良.首先在AOD-Net的第二个特征融合层上添加了第一层的特征图,用全逐点卷积替换了传统卷积方式,并用多尺度结构提升了网络对细节的处理能力.然后用包含有图像重构损失函数、SSIM(Structural similarity)损失函数以及TV(Total variation)损失函数的复合损失函数优化去雾图的对比度、亮度以及色彩饱和度.最后采用分段式的训练方式进一步提升了去雾图的质量.实验结果表明,经该算法去雾后的图像拥有令人满意的去雾结果,图像的饱和度和对比度相较于AOD-Net更自然.与其他对比算法相比,该算法在合成图像实验、真实航拍图像实验以及算法耗时测试的综合表现上更好,更适用于航拍图像实时去雾.展开更多
The spatial structures of China’s Major Function Zoning are important constraining indicators in all types of spatial planning and key parameters for accurately downscaling major functions.Taking the proportion of ur...The spatial structures of China’s Major Function Zoning are important constraining indicators in all types of spatial planning and key parameters for accurately downscaling major functions.Taking the proportion of urbanization zones,agricultural development zones and ecological security zones as the basic parameter,this paper explores the spatial structures of major function zoning at different scales using spatial statistics,spatial modeling and landscape metrics methods.The results show:First,major function zones have spatial gradient structures,which are prominently represented by latitudinal and longitudinal gradients,a coastal distance gradient,and an eastern-central-western gradient.Second,the pole-axis system structure and core-periphery structure exist at provincial scales.The general principle of the pole-axis structure is that as one moves along the distance axis,the proportion of urbanization zones decreases and the proportion of ecological security zones increases.This also means that the proportion of different function zones has a ring-shaped spatial differentiation principle with distance from the core.Third,there is a spatial mosaic structure at the city and county scale.This spatial mosaic structure has features of both spatial heterogeneity,such as agglomeration and dispersion,as well as of mutual,adjacent topological correlation and spatial proximity.The results of this study contribute to scientific knowledge on major function zones and the principles of spatial organization,and it acts as an important reference for China’s integrated geographical zoning.展开更多
在智能监控领域,实现人群计数具有重要价值,针对人群尺度不一、人群密度分布不均及遮挡等问题,提出一种多尺度多任务卷积神经网络(MMCNN)进行人群计数的方法。首先提出一种新颖的自适应人形核生成密度图描述人群信息,消除人群遮挡影响;...在智能监控领域,实现人群计数具有重要价值,针对人群尺度不一、人群密度分布不均及遮挡等问题,提出一种多尺度多任务卷积神经网络(MMCNN)进行人群计数的方法。首先提出一种新颖的自适应人形核生成密度图描述人群信息,消除人群遮挡影响;其次通过构建多尺度卷积神经网络解决人群尺度不一问题,以多任务学习机制同时估计密度图及人群密度等级,解决人群分布不均问题;最后设计一种加权损失函数,提高人群计数准确率。在UCF_CC_50和World Expo'10数据库上进行了评估,验证了自适应人形核的有效性。实验结果表明:所提算法比Sindagi等的方法 (SINDAGI V A,PATEL V M. CNN-based cascaded multi-task learning of high-level prior and density estimation for crowd counting. Proceedings of the 2017 14th IEEE International Conference on Advanced Video and Signal Based Surveillance. Piscataway,NJ:IEEE,2017:1-6)在UCF_CC_50数据库上平均绝对误差(MAE)数值和均方误差(MSE)数值分别降低约1. 7和45;与Zhang等的方法(ZHANG Y,ZHOU D,CHEN S,et al. Single-image crowd counting via multi-column convolutional neural network. Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition. Washington,DC:IEEE Computer Society,2016:589-597)相比,在World Expo'10数据库上所提算法的MAE值降低约1. 5,且在真实公共汽车数据库上仅0～3人的计数误差,表明其实用性较强。展开更多
结合深度学习,提出一种多尺度全卷积神经网络驱动的遥感影像修复方法。构建高斯金字塔实现多尺度变换,影像通过FNEA（fractal net evolution approach）算法进行分割,利用卷积层之间的激活函数值来寻找最佳匹配像素,最后通过全卷积神经...结合深度学习,提出一种多尺度全卷积神经网络驱动的遥感影像修复方法。构建高斯金字塔实现多尺度变换,影像通过FNEA（fractal net evolution approach）算法进行分割,利用卷积层之间的激活函数值来寻找最佳匹配像素,最后通过全卷积神经网络反向传播函数对修补区域进行像素填充。多组实验验证了多尺度全卷积神经网络方法可以良好的修复遥感影像,有自动化高、效率高、目视修复效果佳的优势。展开更多
文摘针对航拍图像易受雾气影响,AOD-Net(All in one dehazing network)算法对图像去雾后容易出现细节模糊、对比度过高和图像偏暗等问题,本文提出了一种基于改进AOD-Net的航拍图像去雾算法.本文主要从网络结构、损失函数、训练方式三个方面对AOD-Net进行改良.首先在AOD-Net的第二个特征融合层上添加了第一层的特征图,用全逐点卷积替换了传统卷积方式,并用多尺度结构提升了网络对细节的处理能力.然后用包含有图像重构损失函数、SSIM(Structural similarity)损失函数以及TV(Total variation)损失函数的复合损失函数优化去雾图的对比度、亮度以及色彩饱和度.最后采用分段式的训练方式进一步提升了去雾图的质量.实验结果表明,经该算法去雾后的图像拥有令人满意的去雾结果,图像的饱和度和对比度相较于AOD-Net更自然.与其他对比算法相比,该算法在合成图像实验、真实航拍图像实验以及算法耗时测试的综合表现上更好,更适用于航拍图像实时去雾.
基金National Natural Science Foundation of China,No.41630644Innovative Think-tank Foundation for Young Scientists of China Association for Science and Technology,No.DXB-ZKQN-2017-048。
文摘The spatial structures of China’s Major Function Zoning are important constraining indicators in all types of spatial planning and key parameters for accurately downscaling major functions.Taking the proportion of urbanization zones,agricultural development zones and ecological security zones as the basic parameter,this paper explores the spatial structures of major function zoning at different scales using spatial statistics,spatial modeling and landscape metrics methods.The results show:First,major function zones have spatial gradient structures,which are prominently represented by latitudinal and longitudinal gradients,a coastal distance gradient,and an eastern-central-western gradient.Second,the pole-axis system structure and core-periphery structure exist at provincial scales.The general principle of the pole-axis structure is that as one moves along the distance axis,the proportion of urbanization zones decreases and the proportion of ecological security zones increases.This also means that the proportion of different function zones has a ring-shaped spatial differentiation principle with distance from the core.Third,there is a spatial mosaic structure at the city and county scale.This spatial mosaic structure has features of both spatial heterogeneity,such as agglomeration and dispersion,as well as of mutual,adjacent topological correlation and spatial proximity.The results of this study contribute to scientific knowledge on major function zones and the principles of spatial organization,and it acts as an important reference for China’s integrated geographical zoning.
文摘在智能监控领域,实现人群计数具有重要价值,针对人群尺度不一、人群密度分布不均及遮挡等问题,提出一种多尺度多任务卷积神经网络(MMCNN)进行人群计数的方法。首先提出一种新颖的自适应人形核生成密度图描述人群信息,消除人群遮挡影响;其次通过构建多尺度卷积神经网络解决人群尺度不一问题,以多任务学习机制同时估计密度图及人群密度等级,解决人群分布不均问题;最后设计一种加权损失函数,提高人群计数准确率。在UCF_CC_50和World Expo'10数据库上进行了评估,验证了自适应人形核的有效性。实验结果表明:所提算法比Sindagi等的方法 (SINDAGI V A,PATEL V M. CNN-based cascaded multi-task learning of high-level prior and density estimation for crowd counting. Proceedings of the 2017 14th IEEE International Conference on Advanced Video and Signal Based Surveillance. Piscataway,NJ:IEEE,2017:1-6)在UCF_CC_50数据库上平均绝对误差(MAE)数值和均方误差(MSE)数值分别降低约1. 7和45;与Zhang等的方法(ZHANG Y,ZHOU D,CHEN S,et al. Single-image crowd counting via multi-column convolutional neural network. Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition. Washington,DC:IEEE Computer Society,2016:589-597)相比,在World Expo'10数据库上所提算法的MAE值降低约1. 5,且在真实公共汽车数据库上仅0～3人的计数误差,表明其实用性较强。
文摘结合深度学习,提出一种多尺度全卷积神经网络驱动的遥感影像修复方法。构建高斯金字塔实现多尺度变换,影像通过FNEA（fractal net evolution approach）算法进行分割,利用卷积层之间的激活函数值来寻找最佳匹配像素,最后通过全卷积神经网络反向传播函数对修补区域进行像素填充。多组实验验证了多尺度全卷积神经网络方法可以良好的修复遥感影像,有自动化高、效率高、目视修复效果佳的优势。
