Seismic 3D Prestack Time Migration on Parallel Computers

Ｓｅｉｓｍｉｃ３ＤＰｒｅｓｔａｃｋＴｉｍｅＭｉｇｒａｔｉｏｎｏｎＰａｒａｌｌｅｌＣｏｍｐｕｔｅｒｓ￥ＬｉｕＰｉｎｇ＆ＣｈｅｎＹｕｎｈｅ（ＷｕｈａｎＤｉｇｉｔａｌＥｎｇｉｎｅｅｒｉｎｇｉｎｓｔｉｔｕｔｅ，ＷｕｃｈａｎｇＰ．Ｏ．Ｂｏｘ７４２２３，Ｐ．Ｒ．...     

Purification of platelet-derived endothelial cell growth inhibitor and its characterization as transforming growth factor-β type 1

In 1986, Brown and Clemmons (Proc. natl Acad. Sci. USA83 (1986) 3321) showed that platelets contain a substance, platelet-derived growth inhibitor (PDGI), that inhibits in vitro endothelial cell replication. Although platelets are rich in transforming grwoth factor (TGF-), PDGI was considered not to be related to TGF-, on the basis of its reported properties (extraction from platelets at neutral pH, binding to heparin-Sepharose). However, we purified PDGI to near homogeneity and showed that on the basis of HPLC retention behavior, in vitro growth inhibitory activities with several cell types, receptor binding, and immunoneutralization of growth inhibitory activity with specific anti-TGF- type 1 antibodies, PDGI is most probably identical with TGF- type 1.     

Flow-induced release of adenosine 5′-triphosphate from endothelial cells of the rat mesenteric arterial bed

Adenosine 5-triphosphate (ATP) was released into the perfusate of rat isolated mesenteric arterial beds during each of two consecutive increases in flow. There was no significant difference between the amounts of ATP released on each occasion. Substance P was also released into the perfusate by increased flow, although its release was more variable. Removal of the endothelium of the mesenteric vessels with sodium deoxycholate led to a significant reduction (74%) in the amount of ATP released compared with the release before the endothelium had been removed. This suggests that the ATP released into the mesenteric arterial perfusate during increased flow arises from endothelial cells.     

Determination of erythrocyte pyrimidine 5′-nucleotidase activity by31P nuclear magnetic resonance: Comparison of normal controls and multiple sclerosis patients

Summary A technique to assay erythrocyte pyrimidine 5-nucleotidase activity in situ using³¹P nuclear magnetic resonance spectroscopy is presented. The assay is chemically specific, simple and applicable to untreated lysates. A comparison of enzyme levels in normal controls and in multiple sclerosis patients employing the assay yielded no significant differences between both groups. Difficulties encountered in the quantitative analysis of the assay using¹H-NMR spectroscopy are briefly discussed.     

基于模糊C均值聚类的二维直流电阻率与射频大地电磁联合反演

为结合直流电阻率（direct current resistivity, DCR）与射频大地电磁（radio-magnetotelluric, RMT）法反演优势,开展了二维DCR与RMT数据联合反演研究。在经典最小结构模型正则化的基础上,采用平衡算子调节两个数据间的权重,引入模糊C均值（fuzzy C-means, FCM）聚类对电阻率模型进行约束,根据数据均方根误差自动调整FCM聚类项的权重,提高了联合反演效果。通过单独反演与联合反演结果的对比,分析了两种方法的反演能力,总结了联合反演的优势。模型试算表明,DCR与RMT数据联合反演得到的电阻率模型较单独反演更接近实际模型,FCM聚类约束的应用可进一步提高联合反演的效果。     

基于电阻率层析扫描的水下抛石体内部空洞探测

为提高水下抛石体的识别精度,构建了预留有内部空洞的抛石体物理模型与电阻率层析扫描模型研究电阻率层析扫描技术在水下抛石体内部空洞探测中的成像效果,并与传统克里金插值法效果进行了对比。结果表明:电阻率层析扫描可以有效识别水下抛石体中的空洞,并且在刺激源数量为6时电阻率层析扫描模型模拟精度达到较高值;电阻率层析扫描技术在通道纵剖面上的识别能力要远高于传统克里金插值法;将数值模拟结果中电导率大于300.μS/cm的区域视为空洞区,整体反演结果略大于真实值,在 x、y、z 3个方向上的偏差分别为2.73、1.75、6.95.cm;基于反演结果对电阻率层析扫描模型进行正演,正演结果与实测数据的拟合度超过0.95,证明了电阻率层析扫描模型反演的准确性。     

基于轻量化深度学习网络的数字仪表识别

数字式仪表常用于变电站、工厂等生产环境,是一种直观的设备监测仪器。然而当前数字式仪表的读取方式还依赖于人工巡检,手动记录等,这些传统的巡检方式来监测设备的运行状态大大降低了巡检效率。为了实现传统行业的数字化转型,本文提出基于轻量化深度学习的数字仪表识别方法,通过改进的YOLOv5的目标检测框架,针对数字仪表目标区域在整张图片大小不一致的情况,提出对于感兴趣区域（ROI）的迭代目标检测方法,首次检测将感兴趣区域进行检测并切割统一到相同的尺度,随后迭代检测网络针对感兴趣区域内的字符进行检测并分类,以达到精确读数的目的。为提升多尺度检测性能,本文采用Res2Net模块主干网络中的的残差模块。采用GIoU取代通用的IoU作为位置损失函数加速模型训练效果的收敛。实验表明,改进后的框架实现了99.62%的准确率和99.55%的召回率,相比基线网络分别提升了12.72%和5.85%。通过将框架在边缘计算平台上的终端部署,在实际生产中取代了人工巡检,实现了商业化运行。     

基于改进YOLOv5的遥感图像飞机目标检测

针对遥感图像中飞机检测尺寸大小不一、背景复杂导致的难以识别问题,提出一种基于YOLOv5网络模型的改进方法。首先,在YOLOv5网络模型中融入Swin-Transformer模块,使网络全局建模并使全维度信息交互,以提升网络的特征提取能力;其次,对损失函数进行优化,引入SIOU损失函数以考虑真实框和预测框之间的向量角度问题。对比实验表明,优化前后网络模型检测精度均为95.3%。在检测精度相同的情况下,改进后的网络模型召回率为91.2%,比改进前提升0.6个百分点;改进后平均检测精度mAP_0.5为95.7%,比改进前提升0.2个百分点。结果表明,改进后的YOLOv5网络模型能在一定程度上提升遥感图像中飞机目标检测性能。     

利用早停可微架构搜索的三维点云模型分类

针对现有三维点云分类网络采用人工设计费时费力的问题,提出早停可微架构搜索(early-stopping differentiable architecture search,ES-DARTS)算法。利用从人工设计网络架构中提取到的先验知识,预定义一个包含高效候选操作的搜索空间,可快速搜索出适用于三维模型分类任务的高性能网络模型;通过追踪网络搜索阶段各候选操作的权重变化,找出跳跃连接操作在双重优化过程中发挥不公平竞争作用的临界点并在此处停止搜索,以保证各候选操作之间的稳定性,解决DARTS算法搜索过程中易出现性能崩溃的问题。提出的算法在ModelNet40数据集上达到了93.2%的识别准确率,比当前人工设计的主流网络具有更高的识别准确率。     

Metabolism of 3β-hydroxycholest-5-en-26-oic acid in hamsters

Summary Metabolism of 26-hydroxycholesterol to 3-hydroxychol-5-en-24-oic acid and other C24-bile acids has been expected to occur by way of 3-hydroxycholest-5-en-26-oic acid in studies in vitro. 3-Hydroxycholest-5-en-26-oic acid was infused intravenously into bile fistula hamsters and the following C24-bile acids were identified: 3-hydroxychol-5-en-24-oic acid, lithocholic acid, chenodeoxycholic acid, and a small amount of cholic acid.