中亚降水数据存在缺失、地理偏差、分辨率低和采集难度大等问题。近年来,神经网络模型被广泛应用于降水降尺度研究。然而,由于山区自然环境复杂多变,普通神经网络模型的预测结果难以解释且适用性差。为此,本文以地理差异分析作为先验知...中亚降水数据存在缺失、地理偏差、分辨率低和采集难度大等问题。近年来,神经网络模型被广泛应用于降水降尺度研究。然而,由于山区自然环境复杂多变,普通神经网络模型的预测结果难以解释且适用性差。为此,本文以地理差异分析作为先验知识约束生成式对抗网络,构建一种新的降水降尺度模型,提高了阿姆河流域复杂环境下降水数据的空间分辨率和精度。首先,依据地形数据通过空间变形模型对输入的Climate Research Units Time Series(CRUTS)降水数据进行空间校正。然后,输入校正后的CRUTS降水数据、气温风速湿度等同化数据及遥感数据到条件生成式对抗网络,重建高分辨率降水数据。最后,考虑到山区降水的各向异性,尤其在地形复杂的上游区域,该模型基于气象站点的真值,对降水数据进行了反距离权重的地理差异分析。结果表明,基于地理差异约束生成式对抗网络的降水降尺度模型能够提升复杂环境降水数据的分辨率和精度。针对中亚阿姆河流域的实验表明,本方法可将CRUTS降水数据的分辨率由55 km提升至11 km,其R2值增加了0.34,均方根误差(RMSE)和平均绝对误差(MAE)分别减小19.4 mm和10.65 mm,偏差(Bias)也由原来的0.24降至0.08。本文为数据采集难、地形地貌复杂区域的降水数据空间分辨率的提高,提供了鲁棒性好、普适性强的方法和思路。展开更多
目的:探讨多排螺旋CT(MDCT)低剂量扫描高分辨率重建在新型冠状病毒肺炎(COVID-19)筛查中的应用价值。方法:选取2020年2月1日~3月1日间就诊的66例疑似COVID-19肺炎患者作为研究对象,将66例患者随机平均分为2组,分别实施胸部常规剂量CT扫...目的:探讨多排螺旋CT(MDCT)低剂量扫描高分辨率重建在新型冠状病毒肺炎(COVID-19)筛查中的应用价值。方法:选取2020年2月1日~3月1日间就诊的66例疑似COVID-19肺炎患者作为研究对象,将66例患者随机平均分为2组,分别实施胸部常规剂量CT扫描(n=33, 120 kV, 300 mAs)和胸部低剂量CT扫描(n=33,100 kV,70 mAs),其中常规剂量组采用512×512矩阵,低剂量组采用1 024×1 024矩阵。同时对胸部低剂量组用4种不同权重的迭代算法进行处理(30%、50%、70%、90%),对比两种检查模式的辐射剂量和图像质量。结果:低剂量组的有效辐射剂量为(1.81±0.14)mSV,与常规剂量组(6.83±0.68)mSV相比降低73.5%(P<0.05);采用1 024大矩阵、90%权重迭代算法的低剂量组图像的CNR、SNR均略低于采用512常规矩阵、90%权重迭代算法的常规剂量组,但差异无统计学意义(SNR:5.11±0.75 vs 5.38±0.41,CNR:5.37±0.33 vs 5.44±0.51, P>0.05);低剂量组患者的肺窗、纵膈窗图像质量主观评分低于常规剂量组,但差异无统计学意义(肺窗:3.30±0.72 vs 3.39±0.78;纵膈窗:3.15±0.90 vs 3.36±0.82, P>0.05)。结论:使用MDCT进行胸部低剂量扫描,同时采用高分辨率重建技术及90%权重迭代算法可用于COVID-19肺炎筛查,可在保证图像质量的前提下显著降低患者所受辐射剂量。展开更多
High-resolution reconstruction of solar speckle image is one of the important research contents in astronomical image processing. High-resolution image reconstruction based on deep learning can obtain the end-to-end m...High-resolution reconstruction of solar speckle image is one of the important research contents in astronomical image processing. High-resolution image reconstruction based on deep learning can obtain the end-to-end mapping function from low-resolution image to high-resolution image through neural network model learning, which can recover the high-frequency information of the image. However, when used to reconstruct the sun speckle image with single feature, more noise and fuzzy local details, there are some shortcomings such as too smooth edge and easy loss of high-frequency information. In this paper, the structure features of input image and reconstructed image are added to CycleGAN network to get MCycleGAN. High frequency information is obtained from structural features by generator network, and the feature difference is calculated to enhance the ability of network to reconstruct high-frequency information. The edge of the reconstructed image is clearer. Compared with the speckle mask method level 1+ used by Yunnan Observatory, the results show that the proposed algorithm has the advantages of small error, fast reconstruction speed and high image clarity.展开更多
China is distinguished by a prominent monsoonal climate in the east of the country, a continental arid climate in the northwest and a highland cold climate on the Qinghai-Tibet Plateau. Because of the long history of ...China is distinguished by a prominent monsoonal climate in the east of the country, a continental arid climate in the northwest and a highland cold climate on the Qinghai-Tibet Plateau. Because of the long history of Chinese civilization, there are abundant and well-dated documentary records for climate variation over the whole of the country as well as many natural archives（e.g., tree-rings, ice cores, stalagmites, varved lake sediments and corals） that enable high-resolution paleoclimatic reconstruction. In this paper, we review recent advances in the reconstruction of climate and extreme events over the last 2000 years in China. In the last 10 years, many new reconstructions, based on multi-proxies with wide spatial coverage, have been published in China. These reconstructions enable us to understand the characteristics of climate change across the country as well as the uncertainties of regional reconstructions. Synthesized reconstructed temperature results show that warm intervals over the last 2000 years occurred in AD 1–200, AD 551–760, AD 951–1320, and after AD 1921, and also show that cold intervals were in AD 201–350, AD 441–530, AD 781–950, and AD 1321–1920. Extreme cold winters, seen between 1500 and 1900, were more frequent than those after 1950. The intensity of regional heat waves, in the context of recent global warming, may not in fact exceed natural climate variability seen over the last 2000 years. In the eastern monsoonal region of China, decadal, multi-decadal and centennial oscillations are seen in rainfall variability. While the ensemble mean for drought/flood spatial patterns across all cold periods shows a meridional distribution, there is a tri-pole pattern with respect to droughts south of 25°N, floods between 25° and 30°N, and droughts north of 30°N for all warm periods. Data show that extreme drought events were most frequent in the periods AD 301–400, AD 751–800, AD 1051–1150, AD 1501–1550, and AD 1601–1650, while extreme flood events were frequent in 展开更多
文摘中亚降水数据存在缺失、地理偏差、分辨率低和采集难度大等问题。近年来,神经网络模型被广泛应用于降水降尺度研究。然而,由于山区自然环境复杂多变,普通神经网络模型的预测结果难以解释且适用性差。为此,本文以地理差异分析作为先验知识约束生成式对抗网络,构建一种新的降水降尺度模型,提高了阿姆河流域复杂环境下降水数据的空间分辨率和精度。首先,依据地形数据通过空间变形模型对输入的Climate Research Units Time Series(CRUTS)降水数据进行空间校正。然后,输入校正后的CRUTS降水数据、气温风速湿度等同化数据及遥感数据到条件生成式对抗网络,重建高分辨率降水数据。最后,考虑到山区降水的各向异性,尤其在地形复杂的上游区域,该模型基于气象站点的真值,对降水数据进行了反距离权重的地理差异分析。结果表明,基于地理差异约束生成式对抗网络的降水降尺度模型能够提升复杂环境降水数据的分辨率和精度。针对中亚阿姆河流域的实验表明,本方法可将CRUTS降水数据的分辨率由55 km提升至11 km,其R2值增加了0.34,均方根误差(RMSE)和平均绝对误差(MAE)分别减小19.4 mm和10.65 mm,偏差(Bias)也由原来的0.24降至0.08。本文为数据采集难、地形地貌复杂区域的降水数据空间分辨率的提高,提供了鲁棒性好、普适性强的方法和思路。
文摘目的:探讨多排螺旋CT(MDCT)低剂量扫描高分辨率重建在新型冠状病毒肺炎(COVID-19)筛查中的应用价值。方法:选取2020年2月1日~3月1日间就诊的66例疑似COVID-19肺炎患者作为研究对象,将66例患者随机平均分为2组,分别实施胸部常规剂量CT扫描(n=33, 120 kV, 300 mAs)和胸部低剂量CT扫描(n=33,100 kV,70 mAs),其中常规剂量组采用512×512矩阵,低剂量组采用1 024×1 024矩阵。同时对胸部低剂量组用4种不同权重的迭代算法进行处理(30%、50%、70%、90%),对比两种检查模式的辐射剂量和图像质量。结果:低剂量组的有效辐射剂量为(1.81±0.14)mSV,与常规剂量组(6.83±0.68)mSV相比降低73.5%(P<0.05);采用1 024大矩阵、90%权重迭代算法的低剂量组图像的CNR、SNR均略低于采用512常规矩阵、90%权重迭代算法的常规剂量组,但差异无统计学意义(SNR:5.11±0.75 vs 5.38±0.41,CNR:5.37±0.33 vs 5.44±0.51, P>0.05);低剂量组患者的肺窗、纵膈窗图像质量主观评分低于常规剂量组,但差异无统计学意义(肺窗:3.30±0.72 vs 3.39±0.78;纵膈窗:3.15±0.90 vs 3.36±0.82, P>0.05)。结论:使用MDCT进行胸部低剂量扫描,同时采用高分辨率重建技术及90%权重迭代算法可用于COVID-19肺炎筛查,可在保证图像质量的前提下显著降低患者所受辐射剂量。
文摘High-resolution reconstruction of solar speckle image is one of the important research contents in astronomical image processing. High-resolution image reconstruction based on deep learning can obtain the end-to-end mapping function from low-resolution image to high-resolution image through neural network model learning, which can recover the high-frequency information of the image. However, when used to reconstruct the sun speckle image with single feature, more noise and fuzzy local details, there are some shortcomings such as too smooth edge and easy loss of high-frequency information. In this paper, the structure features of input image and reconstructed image are added to CycleGAN network to get MCycleGAN. High frequency information is obtained from structural features by generator network, and the feature difference is calculated to enhance the ability of network to reconstruct high-frequency information. The edge of the reconstructed image is clearer. Compared with the speckle mask method level 1+ used by Yunnan Observatory, the results show that the proposed algorithm has the advantages of small error, fast reconstruction speed and high image clarity.
基金Basic Research Project of the Ministry of Science and Technology,No.2011FY120300The "Strategic Priority Research Program" of the Chinese Academy of Sciences,No.XDA05080100Research Project from NSFC,No.41430528
文摘China is distinguished by a prominent monsoonal climate in the east of the country, a continental arid climate in the northwest and a highland cold climate on the Qinghai-Tibet Plateau. Because of the long history of Chinese civilization, there are abundant and well-dated documentary records for climate variation over the whole of the country as well as many natural archives（e.g., tree-rings, ice cores, stalagmites, varved lake sediments and corals） that enable high-resolution paleoclimatic reconstruction. In this paper, we review recent advances in the reconstruction of climate and extreme events over the last 2000 years in China. In the last 10 years, many new reconstructions, based on multi-proxies with wide spatial coverage, have been published in China. These reconstructions enable us to understand the characteristics of climate change across the country as well as the uncertainties of regional reconstructions. Synthesized reconstructed temperature results show that warm intervals over the last 2000 years occurred in AD 1–200, AD 551–760, AD 951–1320, and after AD 1921, and also show that cold intervals were in AD 201–350, AD 441–530, AD 781–950, and AD 1321–1920. Extreme cold winters, seen between 1500 and 1900, were more frequent than those after 1950. The intensity of regional heat waves, in the context of recent global warming, may not in fact exceed natural climate variability seen over the last 2000 years. In the eastern monsoonal region of China, decadal, multi-decadal and centennial oscillations are seen in rainfall variability. While the ensemble mean for drought/flood spatial patterns across all cold periods shows a meridional distribution, there is a tri-pole pattern with respect to droughts south of 25°N, floods between 25° and 30°N, and droughts north of 30°N for all warm periods. Data show that extreme drought events were most frequent in the periods AD 301–400, AD 751–800, AD 1051–1150, AD 1501–1550, and AD 1601–1650, while extreme flood events were frequent in
