This paper proposes a red tide monitoring method based on clustering and modular neural networks. To obtain the features of red tide from a mass of aerial remote sensing hyperspectral data, first the Log Residual Corr...This paper proposes a red tide monitoring method based on clustering and modular neural networks. To obtain the features of red tide from a mass of aerial remote sensing hyperspectral data, first the Log Residual Correction （LRC） is used to normalize the data, and then clustering analysis is adopted to select and form the training samples for the neural networks. For rapid monitoring, the discriminator is composed of modular neural networks, whose structure and learning parameters are determined by an Adaptive Genetic Algorithm （AGA）. The experiments showed that this method can monitor red tide rapidly and effectively.展开更多
Unmanned aerial vehicles(UAV)based remote sensing is an emerging and important data source.Recently,the use of UAVs for remote sensing applications has been rapidly growing owing to their greater availability and the ...Unmanned aerial vehicles(UAV)based remote sensing is an emerging and important data source.Recently,the use of UAVs for remote sensing applications has been rapidly growing owing to their greater availability and the miniaturization of sensors.UAVs are surpassing satellites and aircraft in remote sensing data supply for many local requirements.In comparison with satellite remote sensing data,most UAV remote sensing data is characterized by high resolution,small coverage area,and heterogeneous multi-sources.However,UAVs lack a unified space–time framework and standardized data process.This paper describes a UAV remote sensing data carrier that can be used as an e-commerce platform for data sharing among registered members and a mission planner for new data acquisition.To the best of our knowledge,the data carriers described herein,are the first of their kind.Through seamless docking with UAVs,the data carrier will form a national UAV network,capable of dynamically obtaining very-high-resolution UAV remote sensing images.In practice,a pilot retrieval system of UAV meta data has been developed to provide a catalogue of data product services.展开更多
复杂场景分类是遥感图像解译的一项重要内容。本文通过优化ResNet18深度残差网络和随机森林,实现了遥感图像复杂场景的高精度分类。首先通过数据扩充将数据库扩充以缓解因训练样本少带来的过拟合问题,然后采用ResNet18深度残差网络自动...复杂场景分类是遥感图像解译的一项重要内容。本文通过优化ResNet18深度残差网络和随机森林,实现了遥感图像复杂场景的高精度分类。首先通过数据扩充将数据库扩充以缓解因训练样本少带来的过拟合问题,然后采用ResNet18深度残差网络自动提取遥感图像场景特征,最后使用随机森林分类器实现复杂场景分类任务并分别在NWPU-RESISC45和UC Merced Land Use数据库上进行了实验。结果表明,本文模型场景分类准确率分别为98.86%和99.17%,与单独使用ResNet18深度残差网络相比,本文模型分类准确率分别提高3.36%和1.71%,相比于其他场景分类方法,本文模型分类准确率分别提高5.23%和1.55%。展开更多
针对传统遥感图像数据分析算法存在鲁棒性较差、适应度与计算效率均偏低的问题,文中基于YOLOv3提出了一种轻量化的遥感图像数据分析算法。该算法使用YOLOv3作为神经网络模型的框架,并将内部的Darknet-53多尺度卷积作为主网络。为了减小...针对传统遥感图像数据分析算法存在鲁棒性较差、适应度与计算效率均偏低的问题,文中基于YOLOv3提出了一种轻量化的遥感图像数据分析算法。该算法使用YOLOv3作为神经网络模型的框架,并将内部的Darknet-53多尺度卷积作为主网络。为了减小主网的冗余度,通过SE-Net模型连接网络的深层与浅层卷积,在轻量化的同时也增强了模型的深度特征提取能力。同时,根据改进后网络的权重输出结果,采用剪枝算法对卷积核进行简化,进而完成了模型的轻量化。在实验测试中,轻量化后的模型可显著提升FPS(Frames Per Second)值,且算法的mAP指标为93.25%,在对比算法中为最优,表明了算法模型的有效性及其性能的优越性。展开更多
基金This research was fully supported by the National 863 Natural Science Foundation of P.R.China(2001 AA636030).
文摘This paper proposes a red tide monitoring method based on clustering and modular neural networks. To obtain the features of red tide from a mass of aerial remote sensing hyperspectral data, first the Log Residual Correction （LRC） is used to normalize the data, and then clustering analysis is adopted to select and form the training samples for the neural networks. For rapid monitoring, the discriminator is composed of modular neural networks, whose structure and learning parameters are determined by an Adaptive Genetic Algorithm （AGA）. The experiments showed that this method can monitor red tide rapidly and effectively.
基金Strategic Priority Research Program of Chinese Academy of Sciences(No.XDA19050501)the National Natural Science Foundation of China(grant number 41771388,41971359)。
文摘Unmanned aerial vehicles(UAV)based remote sensing is an emerging and important data source.Recently,the use of UAVs for remote sensing applications has been rapidly growing owing to their greater availability and the miniaturization of sensors.UAVs are surpassing satellites and aircraft in remote sensing data supply for many local requirements.In comparison with satellite remote sensing data,most UAV remote sensing data is characterized by high resolution,small coverage area,and heterogeneous multi-sources.However,UAVs lack a unified space–time framework and standardized data process.This paper describes a UAV remote sensing data carrier that can be used as an e-commerce platform for data sharing among registered members and a mission planner for new data acquisition.To the best of our knowledge,the data carriers described herein,are the first of their kind.Through seamless docking with UAVs,the data carrier will form a national UAV network,capable of dynamically obtaining very-high-resolution UAV remote sensing images.In practice,a pilot retrieval system of UAV meta data has been developed to provide a catalogue of data product services.
文摘复杂场景分类是遥感图像解译的一项重要内容。本文通过优化ResNet18深度残差网络和随机森林,实现了遥感图像复杂场景的高精度分类。首先通过数据扩充将数据库扩充以缓解因训练样本少带来的过拟合问题,然后采用ResNet18深度残差网络自动提取遥感图像场景特征,最后使用随机森林分类器实现复杂场景分类任务并分别在NWPU-RESISC45和UC Merced Land Use数据库上进行了实验。结果表明,本文模型场景分类准确率分别为98.86%和99.17%,与单独使用ResNet18深度残差网络相比,本文模型分类准确率分别提高3.36%和1.71%,相比于其他场景分类方法,本文模型分类准确率分别提高5.23%和1.55%。
文摘针对传统遥感图像数据分析算法存在鲁棒性较差、适应度与计算效率均偏低的问题,文中基于YOLOv3提出了一种轻量化的遥感图像数据分析算法。该算法使用YOLOv3作为神经网络模型的框架,并将内部的Darknet-53多尺度卷积作为主网络。为了减小主网的冗余度,通过SE-Net模型连接网络的深层与浅层卷积,在轻量化的同时也增强了模型的深度特征提取能力。同时,根据改进后网络的权重输出结果,采用剪枝算法对卷积核进行简化,进而完成了模型的轻量化。在实验测试中,轻量化后的模型可显著提升FPS(Frames Per Second)值,且算法的mAP指标为93.25%,在对比算法中为最优,表明了算法模型的有效性及其性能的优越性。
