Direct Measurements of Deep Currents in the Northern Japan Sea

Long-term current measurements by means of subsurface moorings were made for the first time at seven sites in the Japan Basin, the northern part of the Japan Sea. The objective was to directly explore the velocity field in the highly homogeneous deep water mass (the Japan Sea Proper Water) that occupies depths below 500 m. On each mooring three current meters were equipped at an approximately equal distance below about 1000 m depth. Duration of the measurements was 1 to 3 years depending on specific site. This paper describes the basic data set from the moored measurements. It is found that the deep water of the Japan Basin is very energetic with eddies and vertically coherent currents of the order of 0.1 m/s. Surprisingly, the currents and eddies exhibit strong seasonal dependence even in the deepest layers of the Basin. The observed new current features are discussed in comparison with conventional deep circulation pictures derived from hydrographic data. This revised version was published online in August 2006 with corrections to the Cover Date.     

The new approach to global deep sounding

The new approach to global geomagnetic sounding is developed to overcome difficulties of spherical harmonical analysis and subsequent transfer function determination. The approach is based on minimizing the discrepancy between experimental and simulated magnetic fields. The discrepancy is considered as a function of the medium model parameters and the coefficients of external fields. The method can be used for laterally inhomogeneous as well as homogeneous earth models. An example of its application to a radially symmetric model is demonstrated.     

A new theoretical solution of the effect of atmospheric pressure on water level

Based on the partial differential equation governing the effect of atmospheric pressure on water level of confined well, deriving the boundary condition and considering the seepage water between well and aquifer, the author obtained the analytical solution of water level change in time domain under the action of an atmospheric pressure history with the Laplace transform method. This solution is composed of two terms:stable and retarded terms. The stable term is the multiplication of barometric efficiency and simultaneous atmospheric pressure, and it implies the value of water level after infinite time when the atmospheric pressure is a constant from the time in question. The retarded term is the transient process due to the time lag of water exchange between well and aquifer. From the solution, it is obtained that the interference of atmospheric pressure on water level is the integral superimposition of the contribution of all atmospheric pressure changes before the time in question. So that, we further found out the response function of pulsive atmospheric pressure history. Calculation shows: (1) The pulsive response function starts from zero and tends to a steady value, which is proportional to the barometric efficiency, when the time tends to infinity; (2) The retarded time depends on the mechanical property of aquifer and the radius of well. The larger the seepage coefficient, the smaller the radius of well and the thicker the aquifer, then the shorter the retarded time gets. This solution can be used as the theoretical basis for further analysis of the atmospheric effect and practical correcting method in the future.     

多体制遥感卫星成像数据高精度处理新方法

光学和SAR等对地观测卫星需要经过成像、辐射/几何校正等处理和不断的时序积累,才能为计算机解译提供精度高、稳定性好、时间持续的数据和特征。传统中低分辨率对地观测卫星通常基于地理网格内地物目标电磁波散射特性简化为理想点目标的假设,进行逐像素处理。然而,高分宽幅、大斜视、多通道等新体制卫星的工作模式更加复杂,其数据处理对星地全链路各环节产生的误差非常敏感,对成像参数标定或估计的精度提出了更高要求,此时基于理想点目标假设来进行参数估计、成像及校正处理的方式已难以满足处理精度要求。并且,近年来多体制卫星组网式协同和融合应用的新发展,也使得当前的理想点目标假设难以表征和建模多源多时相数据特征。为此,本文提出了多体制遥感卫星成像数据高精度处理的新方法,首先创新提出了“超像素”的概念和表征理论框架,建立了基于超像素的精确成像模型,然后通过挖掘超像素稳定特征并借鉴生成对抗学习机制,实现了星地全链路高耦合成像参数的高精度估计和持续精化,有效提升了多体制遥感卫星成像数据产品的精度,为计算机解译提供了好的数据产品输入。     

多源遥感影像深度识别模型对抗攻击鲁棒性评估

基于深度神经网络的多源遥感影像目标识别系统已逐步在空天遥感情报侦察、无人作战自主环境认知、多模复合末制导等多个军事场景中广泛应用。然而,由于深度学习理论上的不完备性、深度神经网络结构设计工程上的强复用性、以及多源成像识别系统在复杂电磁环境中易受到各类干扰等多因素的影响,使得现有识别系统在对抗攻击鲁棒性方面评估不足,存在极大安全隐患。本文首先从深度学习理论不完备性和识别系统攻击样式两个方面分析了潜在安全风险,并重点介绍了深度识别模型对抗样本攻击基本原理和典型方法。其次,针对光学遥感影像和SAR遥感影像两类典型数据形式,从鲁棒正确识别率和对抗攻击可解释性两个方面开展多源遥感影像深度识别模型对抗攻击鲁棒性评估,覆盖了9类常见深度识别网络架构和7类典型对抗样本攻击方法,验证了现有深度识别模型对抗攻击鲁棒性普遍不足的问题,分析了对抗样本与正常样本的多隐层特征激活差异,为下一步设计对抗样本检测算法和提升模型对抗鲁棒性提供参考。     

车载激光雷达点云快速栅格化算法研究及应用

针对车载激光雷达点云数据量大、密度高且存在分层错位和噪点等情况,提出了一种具实时性激光点云快速栅格化算法,该算法根据雷达扫描精度预设栅格单元大小,可在不丢失对象形状特征的情况下,能快速完成点云数据平滑及降采样处理,并将数据量缩小为处理前的60%。将该栅格算法处理后的点云数据应用于深度学习,作为pointnet++神经网络的训练集及测试集,完成语义分割模型训练与测试。实验结果表明,该算法可在1s内完成上百万量级的点云栅格处理,并且经该算法处理后的点云数据能有效缩短训练时长、提升网络测试精度。     

无人机图像配电导线断股检测的深度学习方法

配电导线的断股易导致断线事故,给用电需要和用电安全带来了极大的负面影响。传统人工检视方法费时费力,而基于无人机（unmannedaerialvehicle,UAV）获取导线影像,利用全卷积网络（fullyconvolutionalnetwork,FCN）深度学习的方法进行配电导线断股快速准确检测可以事半功倍。首先,顾及断股图像空间低占比特点,利用卷积层代替池化层以减少细节损失,实现改进FCN网络架构构建;其次,通过图像变换方法增强原始数据集,提升网络泛化能力;然后,选用BReLU（bilateralrectifiedlinearunit）激活函数,弥补常用激活函数存在的梯度弥散缺陷,提高准确识别率;最后以某电力配电导线巡检项目为例,训练、优化网络,并对UAV获取的图像进行检测,取得了93%的正确率。同时,对比分析了该方法与传统方法的检测效果,结果表明了在配电导线断股智能识别中该方法的鲁棒性和准确性显著占优。     

基于金字塔瓶颈残差网络优化算法的多光谱影像分类

提出了一种使用深度金字塔残差网络来构造分类网络的方法,用于多光谱影像分类。该方法利用残差模块加深网络复杂度,利用金字塔瓶颈模块提取图像中的高层特征。针对高分一号、RapidEye和GeoEye卫星影像的实验结果表明,新方法表现出优越的泛化能力和更高的精度。     

Coupled discrete element and finite volume solution of two classical soil mechanics problems

One dimensional solutions for the classic critical upward seepage gradient/quick condition and the time rate of consolidation problems are obtained using coupled routines for the finite volume method (FVM) and discrete element method (DEM), and the results compared with the analytical solutions. The two phase flow in a system composed of fluid and solid is simulated with the fluid phase modeled by solving the averaged Navier–Stokes equation using the FVM and the solid phase is modeled using the DEM. A framework is described for the coupling of two open source computer codes: YADE-OpenDEM for the discrete element method and OpenFOAM for the computational fluid dynamics. The particle–fluid interaction is quantified using a semi-empirical relationship proposed by Ergun [12]. The two classical verification problems are used to explore issues encountered when using coupled flow DEM codes, namely, the appropriate time step size for both the fluid and mechanical solution processes, the choice of the viscous damping coefficient, and the number of solid particles per finite fluid volume.     

鄂尔多斯盆地北部盒8段古含气范围的地球化学特征

鄂尔多斯盆地盒8段气藏为深盆气藏。通过分析鄂尔多斯盆地苏里格庙地区盒8段储集层中有机包裹体的丰度、储集层的热解特征,并结合该区的构造演化和生烃演化等特征,探讨了苏里格庙北部盒8段气层的古含气边界的变化情况及天然气的保存条件。认为盒8段气层的分布范围在早白垩世末期达到最大,后期的构造抬升对气水边界的影响并不大,这是因为构造抬升后,气源岩的压力下降,增加了煤系地层中吸附气的脱附能力,为气藏补充足够的气体,从而使得古气水边界的分布与现今的差别不大。地层倾角的大小能影响天然气的富集和保存,地层倾角相对平缓的斜坡部位利于天然气的聚集,是高产气区。