气象观测资料表明,在当前全球暖化背景下高海拔地区的增温速率更快,其气候对全球气候变化的响应更为敏感。但是,高海拔地区如何响应快速气候事件,由于古气候古环境重建资料在高海拔地区较少,这一问题还没有很好的回答。为此选择云南省高山湖泊错恰湖为研究对象(海拔约3960 m),分析了湖泊沉积岩芯(深度范围90~244 cm)中正构烷烃的分布特征,重建了末次冰消期(19000~9500 a B.P.)沉积物有机质来源的变化,进而推断古气候演变。在17800~17000 a B.P.,错恰湖有机质以陆源输入为主,水生贡献相对减少,气候以暖湿为主;在17000~15100 a B.P.,水生有机质的贡献的比例增加,气候以冷干为主;在15100~12700 a B.P.,湖泊沉积有机质的陆源贡献增加,水生贡献相对减少,气候相对暖湿;在12700~11400 a B.P.时段,湖泊沉积水生有机质来源相对增多,气候相对冷干。与其他区域和全球气候记录对比发现,错恰湖的沉积记录指示的气候变化事件,受高纬冰量以及北大西洋驱动的西南季风突变所影响,在末次冰消期记录的4个明显的千年尺度气候事件,在时间上与First warmth、H1、B-A和YD事件相对应。与云南地区其他湖泊记录对比发现,小型湖泊或高海拔湖泊对这些全球快速气候事件的响应更加敏感。
针对高光谱影像分类中的深度学习模型设计问题,提出了一种面向高光谱影像分类的网络结构自动搜索方法。该方法首先利用可微分结构搜索技术在源高光谱数据集上进行网络结构搜索,然后采用堆叠单元的形式构建深度网络模型,最后利用目标高光谱影像对模型进行分类性能评估。该方法仅在源高光谱数据集上进行一次网络结构搜索,得到的深度网络模型即可应用于其他目标高光谱影像的分类任务,能够有效提高模型利用率。为了提高自动搜索得到的模型的泛化能力和分类精度,采用多源多分辨率的高光谱影像构建源数据集,并引入部分通道连接操作提高搜索效率。试验表明,该方法能够自动搜索出适合高光谱影像分类任务且具备一定通用性的深度网络模型,该模型能够取得较常规深度学习模型更为优异的分类效果,在University of Pavia、Indian Pines、Salinas和Houston 2018这4个目标高光谱影像上分别取得了98.15%、98.74%、97.30%和74.47%的总体分类精度。
Inter-satellite links improve the performance of global navigation satellite systems (GNSSs) via communication and ranging. When there are limited facilities and links, a key challenge involves assigning links for the downlink of telemetry data in time and effective ranging. We describe this problem and propose a corresponding link scheduling method consisting of three steps. A genetic algorithm was used to identify optimal downlink routes for all non-visible satellites in the first step. The optimization of minimizing timeslot delays usually has multiple optimal routes. The optimal routes were selected to expand to a superframe’s length, which constituted a downlink route scheme for the superframe. The position dilution of precision (PDOP) of ranging links was limited through satellite selection in the second step. Four visible satellites with minimal PDOP were selected and scheduled with idle timeslots. The PDOP decreased with increasing links, and its upper limit was therefore determined by the selected satellites. The final step was to schedule idle timeslots for visible satellites unless the requested link number was met. To test the feasibility of the proposed method, the link assignment was implemented for 10,080 superframes of a typical GNSS constellation. The final link assignment enabled all satellites to transmit telemetry data back to the facility with a delay of no more than 4 timeslots (with one ground facility tracking), and more than 10 links were obtained with PDOPs approaching the minimum. These scheduling results confirm the utility of the method. 相似文献