特征图强化网络：利用特征图强化船舶检测模型训练的网络结构

随着人工智能技术的进一步发展,深度学习方法在船舶检测领域发挥着重要作用。然而,深度学习算法出现的虚警和漏检,对船舶检测领域技术的应用存在一定的阻碍。虽然经典的深度学习方法能够有效处理单一背景的海面,但是当面对复杂背景之下的数据时,经典模型很容易得出岸上的虚警。并且在常规训练中,模型常常对一些显著特征过于关注,出现特征过拟合现象,当这些显著特征发生改变时极易出现漏检。在模型对输入进行前向传播的过程中,模型中不同网络层会对输入生成对应的映射,也就是特征图。充分利用特征图的语义信息和空间信息是一种有效减少虚警和漏检的方法。与传统模型相比,我们提出的特征图强化网络可以充分利用特征图生成自适应特征图掩码与水陆分割掩码,避免模型的特征过拟合与削弱复杂背景造成的影响,最终达到减少虚警与漏检的目的。在现有公共数据集上与算法模型的对比实验结果表明,本文所提出的方法的性能更为出色,超过了其他SOTA算法。     

GIS科普资源发布系统实现的关键技术

科普资源发布系统利用GIS强大的空间查询与分析功能,方便市民浏览、查询与统计科普资源情况,掌握更多科普知识.本文主要论述了系统设计与实现过程中应用的关键技术,详细给出了栅格影像自动选择的方法,矢量栅格匹配的原理,以及统计图的绘制技术.最后介绍了系统的开发模式与软件环境,给出了统计效果图.     

Demographic Transition: Threat to Developing Nations

Abstract

As computers become commonplace in the nation's schools, interactive maps are finding their way into the geography classroom. Students can use electronic atlases to access information as well as use authoring software to produce their own interactive maps. Does this change in media correspond to better geography education for our children? This article addresses whether using interactive maps for gathering and presenting information can have a positive influence on learning geography content. A two-week experiment was conducted in five 7th grade classrooms. Students in four of the classes researched and produced map products using both traditional and computer-assisted techniques. The fifth class was a control group. A pretest and two post-tests were given to quantitatively assess student learning after each project. Also, students were given a qualitative questionnaire to determine their attitude towards the different techniques. Quantitative results of the study indicate that there was no significant difference in student post-test performance after using each of the techniques, but qualitative results suggest that students prefer using the computer techniques over the traditional methods for gathering and presenting information.     

Using Web Maps to Analyze the Construction of Global Scale Cognitive Maps

Game-based Web sites and applications are changing the ways in which students learn the world map. In this study, a Web map-based digital learning tool was used as a study aid for a university-level geography course in order to examine the way in which global scale cognitive maps are constructed. A network analysis revealed that clicks were negatively correlated with familiarity, but were positively correlated with area. More significantly, well-known countries did not act as central nodes within network-based mental models, but served a more complex mediating role in structuring the space between subregions.     

《上海市地图集》世博专版的设计特点

在《上海市地图集》(中国2010年上海世博会专版)设计与编制实践中,针对《上海市地图集》(以下简称《图集》)的编制背景和上海地域特色,突破了传统城市地图集的内容结构与表达模式,对该图集的内容框架、表示方法运用以及图形效果进行了富有特色的创新设计,使之成为世博会文化工程的亮点之一,获得了国内外一致好评。     

IMP 8 magnetosheath field comparisons with models

This study presents cross-sectional vector maps of the magnetic field derived from IMP 8 magnetometer in the magnetosheath at 30 R_e behind the Earth. In addition the vector patterns of the magnetosheath field for northward, southward, and east-west interplanetary magnetic field (IMF) directions are qualitatively compared with those obtained from the Spreiter-Stahara gas dynamic (GD) and Fedder-Lyon magnetohydrodynamic models (MHD). The main purpose is to display the cross-sectional differences in relation to the dayside merging with different IMF directions, allowing the reader to make direct visual comparisons of the vector patterns. It is seen that for east-west IMF directions, the data-based and MHD-based patterns differ noticeably in a similar way from the GD model, presumably reflecting the influence of dayside magnetic merging of the Earths magnetic field with the y-component of the interplanetary magnetic field. All three northward IMF cross sections show comparable field draping patterns as expected for a closed magnetosphere. For southward IMF case, on the other hand, differences between the three cross-sectional patterns are greater as seen in the field vector sizes and directions, especially closer to the magnetopause where more disturbed magneto-spheric conditions are known to be exist. The data comparisons with the MHD and GD models show that the differences result from the magnetic field-flow coupling and that the effects of dayside reconnection are present in IMP 8 magnetic field measurements.     

Assessment of Erythemal UV Level in Nepal Based on Solar UV Estimates from Total Ozone Mapping Spectrometer

Nepal lies on the southern slope of Himalaya in Asia. In a width ranging between 150 and 250 km, the altitude varies greatly from about 100 m at its southern border to a maximum of 8848 min the northern part. Like the variation in altitude, climatic condition varies quite a lot. Long-term monthly mean erythemal UV daily dose values for Nepal are evaluated using Total Ozone Mapping Spectrometer (TOMS) estimation from the time of its overpass between 1996 and 2003. The results are presented as summer and winter maps of mean UV levels in each satellite grid. The mean winter erythemal UV daily dose ranges between 2.1 and 3.6 kJ m-2 whereas summer values are found to lie between 4.6 and 9.7 kJ m-2. The altitude variation increases the UV levels by about 0.2 kJ km-1 in winter months, and 0.9 kJ km-1 in summer. A multiyear monthly average erythemal daily dose in most of the areas shows that the summer value is about three times higher than that in winter. Although year-to-year variation is not pronounced in high- and mid-elevation regions, UV levels seemed to decrease from 1997 to 2002 in the southern part of the country in the low elevation region by about 5.35%. Due to the combined effects of the altitude, low ozone concentration in the troposphere, and thin air, surface UV radiation at higher altitudes is found to be higher than in the surrounding regions.     

Measured and perceived visual complexity: a comparative study among three online map providers

We present a study on human perception of map complexity, with the objective of better understanding design decisions that may lead to undesirable levels of complexity in web maps. We compare three complexity metrics to human ratings of complexity obtained through a user survey. Specifically, we use two algorithmic approaches published by others, which measure feature congestion (FC) and subband entropy (SE), as well as our own approach of counting object types rather than individual objects. We compare these metrics with each other as well as with human complexity ratings for three maps of the same area from map providers Google Maps, Bing Maps, and OpenStreetMap. Each map design is assessed at three different scales (levels of detail). We find that (1) the FC and SE metrics appear to be adequate predictors of what humans consider complex; (2) object-type counts are slightly less successful at predicting human-rated complexity, implying that clutter is more important in perceived complexity than diversity of symbology; and (3) generalization choices do impact human complexity ratings. These findings contribute to our understanding of what makes a map complex, with implications for designing maps that are easy to use.