Video snapshot compressive imaging(Video SCI) modulates scenes using various encoding masks and captures compressed measurements with a low-speed camera during a single exposure. Subsequently, reconstruction algorithm...Video snapshot compressive imaging(Video SCI) modulates scenes using various encoding masks and captures compressed measurements with a low-speed camera during a single exposure. Subsequently, reconstruction algorithms restore image sequences of dynamic scenes, offering advantages such as reduced bandwidth and storage space requirements. The temporal correlation in video data is crucial for Video SCI, as it leverages the temporal relationships among frames to enhance the efficiency and quality of reconstruction algorithms, particularly for fast-moving objects.This paper discretizes video frames to create image datasets with the same data volume but differing temporal correlations. We utilized the state-of-the-art(SOTA) reconstruction framework, EfficientSCI++, to train various compressed reconstruction models with these differing temporal correlations. Evaluating the reconstruction results from these models, our simulation experiments confirm that a reduction in temporal correlation leads to decreased reconstruction accuracy. Additionally, we simulated the reconstruction outcomes of datasets devoid of temporal correlation, illustrating that models trained on non-temporal data affect the temporal feature extraction capabilities of transformers, resulting in negligible impacts on the evaluation of reconstruction results for non-temporal correlation test datasets.展开更多
Ultrahigh-speed imaging is an essential tool for capturing fast dynamic scenes across various fields.Despite the development of numerous technical strategies,achieving ultrahigh-speed imaging with high spatiotemporal ...Ultrahigh-speed imaging is an essential tool for capturing fast dynamic scenes across various fields.Despite the development of numerous technical strategies,achieving ultrahigh-speed imaging with high spatiotemporal resolution and substantial sequence depth remains a significant challenge.To address this issue,we present a compressive ultrahigh-speed imaging technique based on acousto-optic frequency sweeping,termed AOFSCUSI.AOFS-CUSI employs light with rapidly time-varying spectra generated by acousto-optic modulation to illuminate dynamic scenes,records spatio-spectral information using snapshot compressive imaging,and ultimately reconstructs spatiotemporal information through time-spectrum mapping.This technique achieves a temporal resolution of 1.55 million frames per second,a spatial resolution of 228 lp/mm,and a sequence depth of 31 in a single shot.We experimentally validate the superior performance of AOFS-CUSI by capturing the rotation of an optical chopper,the movement of microspheres in a microchannel,and the femtosecondlaser-induced cavitation bubble dynamics.By eliminating the requirement for ultrafast laser sources and simultaneously extending the temporal window,AOFS-CUSI offers an excellent solution for recording and analyzing various fast dynamics,presenting significant potential for applications in both fundamental and applied research.展开更多
针对正交频分线性调频(OFD-LFM)信号MIMO高分辨雷达稀疏成像问题展开研究,在分析了OFD-LFM信号频谱合成原理以及MIMO高分辨雷达一次快拍成像原理的基础上,给出了一种基于频域稀疏OFD-LFM信号和空域稀疏MIMO雷达天线阵列的联合稀疏模型,...针对正交频分线性调频(OFD-LFM)信号MIMO高分辨雷达稀疏成像问题展开研究,在分析了OFD-LFM信号频谱合成原理以及MIMO高分辨雷达一次快拍成像原理的基础上,给出了一种基于频域稀疏OFD-LFM信号和空域稀疏MIMO雷达天线阵列的联合稀疏模型,并结合压缩感知理论,提出了目标高分辨距离像(high-resolution range profile,HRRP)合成方法以及目标二维成像方法。该方法能够在大幅减少OFD-LFM信号子载波个数、大幅减少MIMO高分辨雷达天线阵元个数的条件下,利用一次快拍重构出高质量的目标HRRP和二维像,不仅避免了目标机动带来的运动补偿难题,同时还有利于天线阵列的工程实现。仿真结果表明所提方法是有效的,且具有一定的抗噪性。展开更多
准确的云分类模型对气象监测有重要的意义,传统机器学习云分类模型依赖手工特征提取,容易受噪声数据影响,模型泛化能力较差。深度网络分类模型能自动学习图像深度特征,但是对于图像边缘与细节分类效果不佳。本文针对上述问题进行研究。...准确的云分类模型对气象监测有重要的意义,传统机器学习云分类模型依赖手工特征提取,容易受噪声数据影响,模型泛化能力较差。深度网络分类模型能自动学习图像深度特征,但是对于图像边缘与细节分类效果不佳。本文针对上述问题进行研究。首先提取Himawari-8卫星云图光谱特征、纹理特征用以训练模糊支持向量机(Fuzzy Support Vector Machine,FSVM)模型;同时利用不同通道云图训练深度网络,学习云图深度特征;最后,根据不同模型特性,训练元分类器对各模型输出进行融合,设计了一种基于深度网络与FSVM集成学习的云分类方法,该方法综合不同模型优势,利用不同模型间的互补性提高云分类结果的鲁棒性和可信度。相比单独使用FSVM或深度网络的分类模型,本文集成学习方法在众多评价指标中有更好的表现,平均命中率、平均误报率和平均临界成功指数分别达到0.9245、0.0796、0.8581;与其它云分类模型相比,本文方法也有更好的分类效果;在具体案例测试中也发现,该方法对于不同云类混合区有更高的识别精度,而且能更加准确的识别云团边缘及细节。本文模型能够满足云分类模型稳定可靠、高精度、泛化性能强的要求。展开更多
基金supported in part by the National Natural Science Foundation of China (No. U23B2011)。
文摘Video snapshot compressive imaging(Video SCI) modulates scenes using various encoding masks and captures compressed measurements with a low-speed camera during a single exposure. Subsequently, reconstruction algorithms restore image sequences of dynamic scenes, offering advantages such as reduced bandwidth and storage space requirements. The temporal correlation in video data is crucial for Video SCI, as it leverages the temporal relationships among frames to enhance the efficiency and quality of reconstruction algorithms, particularly for fast-moving objects.This paper discretizes video frames to create image datasets with the same data volume but differing temporal correlations. We utilized the state-of-the-art(SOTA) reconstruction framework, EfficientSCI++, to train various compressed reconstruction models with these differing temporal correlations. Evaluating the reconstruction results from these models, our simulation experiments confirm that a reduction in temporal correlation leads to decreased reconstruction accuracy. Additionally, we simulated the reconstruction outcomes of datasets devoid of temporal correlation, illustrating that models trained on non-temporal data affect the temporal feature extraction capabilities of transformers, resulting in negligible impacts on the evaluation of reconstruction results for non-temporal correlation test datasets.
基金National Natural Science Foundation of China(12325408,12274129,12374274,12274139,62175066,92150102,62475070,12474404,12471368)Shanghai Municipal Education Commission(2024AI01007)+1 种基金Open Fund of Guangdong Provincial Key Laboratory of Nanophotonic Manipulation(202504)Guangdong ST Program(2023B1212010008).
文摘Ultrahigh-speed imaging is an essential tool for capturing fast dynamic scenes across various fields.Despite the development of numerous technical strategies,achieving ultrahigh-speed imaging with high spatiotemporal resolution and substantial sequence depth remains a significant challenge.To address this issue,we present a compressive ultrahigh-speed imaging technique based on acousto-optic frequency sweeping,termed AOFSCUSI.AOFS-CUSI employs light with rapidly time-varying spectra generated by acousto-optic modulation to illuminate dynamic scenes,records spatio-spectral information using snapshot compressive imaging,and ultimately reconstructs spatiotemporal information through time-spectrum mapping.This technique achieves a temporal resolution of 1.55 million frames per second,a spatial resolution of 228 lp/mm,and a sequence depth of 31 in a single shot.We experimentally validate the superior performance of AOFS-CUSI by capturing the rotation of an optical chopper,the movement of microspheres in a microchannel,and the femtosecondlaser-induced cavitation bubble dynamics.By eliminating the requirement for ultrafast laser sources and simultaneously extending the temporal window,AOFS-CUSI offers an excellent solution for recording and analyzing various fast dynamics,presenting significant potential for applications in both fundamental and applied research.
文摘针对正交频分线性调频(OFD-LFM)信号MIMO高分辨雷达稀疏成像问题展开研究,在分析了OFD-LFM信号频谱合成原理以及MIMO高分辨雷达一次快拍成像原理的基础上,给出了一种基于频域稀疏OFD-LFM信号和空域稀疏MIMO雷达天线阵列的联合稀疏模型,并结合压缩感知理论,提出了目标高分辨距离像(high-resolution range profile,HRRP)合成方法以及目标二维成像方法。该方法能够在大幅减少OFD-LFM信号子载波个数、大幅减少MIMO高分辨雷达天线阵元个数的条件下,利用一次快拍重构出高质量的目标HRRP和二维像,不仅避免了目标机动带来的运动补偿难题,同时还有利于天线阵列的工程实现。仿真结果表明所提方法是有效的,且具有一定的抗噪性。
文摘准确的云分类模型对气象监测有重要的意义,传统机器学习云分类模型依赖手工特征提取,容易受噪声数据影响,模型泛化能力较差。深度网络分类模型能自动学习图像深度特征,但是对于图像边缘与细节分类效果不佳。本文针对上述问题进行研究。首先提取Himawari-8卫星云图光谱特征、纹理特征用以训练模糊支持向量机(Fuzzy Support Vector Machine,FSVM)模型;同时利用不同通道云图训练深度网络,学习云图深度特征;最后,根据不同模型特性,训练元分类器对各模型输出进行融合,设计了一种基于深度网络与FSVM集成学习的云分类方法,该方法综合不同模型优势,利用不同模型间的互补性提高云分类结果的鲁棒性和可信度。相比单独使用FSVM或深度网络的分类模型,本文集成学习方法在众多评价指标中有更好的表现,平均命中率、平均误报率和平均临界成功指数分别达到0.9245、0.0796、0.8581;与其它云分类模型相比,本文方法也有更好的分类效果;在具体案例测试中也发现,该方法对于不同云类混合区有更高的识别精度,而且能更加准确的识别云团边缘及细节。本文模型能够满足云分类模型稳定可靠、高精度、泛化性能强的要求。
