基于用户点击数据的细粒度图像识别方法概述

近年来,细粒度图像识别逐渐成为计算机视觉领域的研究热点.由于不同类别图像间的视觉差异小、语义鸿沟问题严重,传统的基于视觉特征的细粒度图像识别性能往往不尽人意.针对这些挑战,目前许多学者都在研究基于用户点击数据的图像识别.本文围绕点击数据在图像识别中数据预处理、特征提取和模型构建3大模块中的应用,总结了已有的基于点击数据的识别算法及最新的研究进展.     

一种闭合条件约束的全局最优多视点云配准方法

针对已有多视点云配准存在的问题,提出了一种严密的闭合条件约束配准方法。该方法首先采用"点-切平面"迭代最近点算法分别求解各对应点云间的坐标转换参数;再以单站点云为配准单元,并将转换参数视为随机观测值构建条件方程,采用条件平差方法对转换参数作改正以达到全局最优。通过地面三维激光扫描仪实测两组点云数据进行试验,验证了该方法有效且可行。试验结果表明,对采样间隔为毫米级和厘米级的点云,增加扫描重叠度能够提高配准精度和可靠性。     

重力场模型及GNSS/水准的区域似大地水准面精化

针对现有地球重力场模型综合利用研究较少的情况,该文利用实测全球导航卫星系统/水准数据分析GOCO03S、ITG-GRACE2010S、GO_CONS_GCF_2_DIR_R4、GO_CONS_GCF_2_TIM_R4和EGM2008等地球重力场模型不同谱域位系数对应的高程异常精度,提出对多类重力场模型进行简单谱组合和加权谱组合,并进行精度分析;然后利用这两类组合重力场模型,结合全球导航卫星系统/水准数据对区域似大地水准面的精化展开研究。计算结果表明:在实验区域,与EGM2008模型相比,采用简单谱组合法和加权谱组合法均能提高模型高程异常的精度,标准差最优分别可达0.081m和0.084m,对应精度提高幅度分别为40.9%和38.5%;以多类重力场模型为基础,经简单谱组合法或加权谱组合法得到组合重力场模型,并利用全球导航卫星系统/水准数据进行精化,可获得较高精度的区域似大水准面,精度最优可达0.048m。     

径向基神经网络在沉降预测中的应用

针对大型建筑物的沉降监测存在较多的不等时间间隔沉降监测数据,现有模型需要对此类数据进行等间隔处理后建立沉降预测模型的现状,该文采用无需对监测数据进行等间隔预处理的径向基神经网络对沉降监测数据建立沉降预测。通过对西安某大厦基坑开挖对地表和周围建筑物影响的沉降监测数据进行实例分析,并与非等间隔灰色GM(1,1)预测模型进行对比,利用模型评价指标评价预测模型精度。结果表明:采用径向基神经网络建立预测模型处理过程简便,其预测精度优于非等间隔灰色GM(1,1)预测模型。     

Spatio-Temporal Characteristics of Heavy Precipitation Forecasts fromECMWF in Eastern China

This study examines the spatio-temporal characteristics of heavy precipitation forecasts in eastern China fromthe European Centre for Medium-Range Weather Forecasts (ECMWF) using the time-domain version of the Method forObject-based Diagnostic Evaluation (MODE-TD). A total of 23 heavy rainfall cases occurring between 2018 and 2021 areselected for analysis. Using Typhoon “Rumbia” as a case study, the paper illustrates how the MODE-TD method assessesthe overall simulation capability of models for the life history of precipitation systems. The results of multiple tests withdifferent parameter configurations reveal that the model underestimates the number of objects’ forecasted precipitationtracks, particularly at smaller radii. Additionally, the analysis based on centroid offset and area ratio tests for differentclassified precipitation objects indicates that the model performs better in predicting large-area, fast-moving, and long lifespan precipitation objects. Conversely, it tends to have less accurate predictions for small-area, slow-moving, and short lifespan precipitation objects. In terms of temporal characteristics, the model overestimates the forecasted movement speedfor precipitation objects with small-area, slow movement, or both long and short lifespans while underestimating it forprecipitation with fast movement. In terms of temporal characteristics, the model tends to overestimate the forecastedmovement speed for precipitation objects with small-area, slow movement, or both long and short lifespans while un derestimating it for precipitation with fast movement. Overall, the model provides more accurate predictions for theduration and dissipation of precipitation objects with large-area or long-lifespan (such as typhoon precipitation) whilehaving large prediction errors for precipitation objects with small-area or short-lifespan. Furthermore, the model’s simu lation results regarding the generation of precipitation objects show that it performs relatively well in simulating thegeneration of large-area and fast-moving precipitation objects. However, there are significant differences in the forecastedgeneration of small-area and slow-moving precipitation objects after 9 hours.     

海洋桡足类氨基酸组成及与饵料和光照的关系

桡足类氨基酸含量受其饵料浮游植物的影响。同一种桡足类因采集地不同，氨基酸的总量也不相同；19种氨基酸重量百分比随不同桡足类而异，但是不同桡足类各种氨基酸占总氨基酸的比例除胱氨酸和牛磺酸的变化比较大以外，其它都基本相同；胱氨酸和牛磺酸的变化与总氨基酸不成正比。各海区不同植食性桡足类的氨基酸组成和海区浮游植物氨基酸的组成有着很好的相关性，相关系数R^2在0．85以上。桡足类胱氨酸占总氨基酸的比例与采样站海水1％光衰减深度有很好的相关关系，其回归方程为y=0．28091nx 0．131，R^2=0．6427，n=13；随海水1％光衰减深度的增加，胱氨酸的含量以对数形式递增。桡足类氨基酸与UV的关系有待进一步研究。     

桶形基础平台海上沉放安装

桶形基础平台是世界上一种最新型的海洋平台，可广泛应用于海上油气田开发。为满足我国际油气田开发中卫星平台的结构和功能要求，桶形基础平台采用新的压载沉放方法安装，具有更高的效率和经济性。沉放过程中桶形基础平台受到水流和波浪的作用，结合模型试验对平台运动进行数值模拟分析，以了解在整个沉放过程中相应参数的变化趋势。根据桶形基础平台的特征，制定联台沉放方案，并通过压载实现平台的沉放控制。     

码头系泊船舶模型试验

通过对一艘50kDWG码头系泊船舶的模型试验，得出了系泊缆绳张力随外载荷变化的一般规律，对码头系泊有一定的参考价值。     

A Finite Element Solution of Wave Forces on Submerged Horizontal Circular Cylinders

In this paper, a numerical model is established for estimating the wave forces on a submerged horizontal circular cylinder. For predicting the wave motion, a set of two-dimensional Navier-Stokes equations is solved numerically with a finite element method. In order to track the moving non-linear wave surface boundary, the Navier-Stokes equations are discretized in a moving mesh system. After each computational time step, the mesh is modified according to the changed wave surface boundary. In order to stabilize the numerical procedure, a three-step finite element method is applied in the time integration. The water sloshing in a tank and wave propagation over a submerged bar are simulated for the first time to validate the present model. The computational results agree well with the analytical solution and the experimental data.Finally, the model is applied to the simulation of interaction between waves and a submerged horizontal circular cylinder.The effects of the KC number and the cylinder depth on the wave forces are studied.     

Trial and Numerical Analysis of Specimen Pipelay

Subsea pipelay has a relatively long history. In recent years, there has been a domestic need for the laying of large diameter thin wall pipes. A land-based trial for the large diameter thin wall specimen pipe is described in this paper. Re-gression analysis is performed for the trial data and the formula derived can express the trial data very well. Numerical analysis is adopted to compute various trial conditions. Then the numerical model is revised with the trial results, which are consistent with each other. After summarization of the results of trial and numerical analysis, the characteristics are described of the spatial configuration during the laying of the pipe and it is concluded that the maximum strain appears around the center of the raised pipeline. In the end, a reference Standard, which limits the maximum stress below the yield stress is provided.