一种多层特征融合的道路场景实时分割方法     

周继苗  李必军  陈世增 《测绘通报》2020,(1):10-15

道路场景理解是自动驾驶领域中重要模块之一，它可以提供关于道路更丰富的信息，对于建立高精度地图和实时规划都具有重要作用。其中，语义分割可以为图像每个像素赋予类别信息，是自动驾驶场景理解中最常用的方法。但是，目前常用的语义分割算法在速度和精度上大都不能达到很好的平衡。本文在MobileNetV2的基础上，提出了一种多层次特征融合的方法，使得网络可以在实时运行的同时保证精度满足实际应用的需求，并在Cityscapes数据集上进行了试验验证和分析。  相似文献   

复杂卫星图像中的小目标船舶识别   总被引：1，自引：1，他引：0  

姚红革  王诚  喻钧  白小军  李蔚 《遥感学报》2020,24(2):116-125

船舶作为海上的重要目标,实现对船舶自动识别有重要的意义。针对卫星图像中云雾、海岸背景等复杂海情对船舶识别带来的干扰,以及小目标船舶高漏检率问题,本文提出一种多尺度深度学习模型训练策略,在此基础上构建了一种船舶识别的深度学习网络,该网络可分为多尺度训练、特征提取、生成目标建议区域、船舶分类这4个部分。首先,采用多尺度的训练策略,将多尺度的船舶样本送入网络中进行训练,这样在训练样本中加入了大量小目标船舶的样本,使网络充分提取到小目标船舶的特征;其次,通过卷积神经网络对目标船舶进行特征自适应提取;然后,目标区域建议网络可依据卷积神经网络提取到的特征,在图像中找到感兴趣目标区域,即框定船舶的位置;最后,通过多个全连接层的组合,将高维特征映射到一个4元组中,再运用分类函数输出每一类船舶的概率值,概率值最大的则为该船舶的类别。同时为解决云雾遮挡和海岸背景的干扰,采用了一种负样本增强学习的方法,在样本数据集中加入了大量只含有云雾和海岸背景的图片,进行负样本扩充,增强网络模型对云雾及海岸背景的特征学习能力,以此解决复杂海情的影响。实验结果表明,所提方法有效解决了复杂海情条件下的船舶识别难,以及小目标船舶识别难的问题,实现了复杂海情条件下的船舶识别。同时,与现有成熟的深度学习目标识别算法相比,本文算法的精确度和召回率分别提升了6.98%和18.17%,所训练的模型具有良好的泛化能力和鲁棒性。  相似文献   

Macrobenthic infaunal communities associated with deep‐sea hydrocarbon seeps in the northern Gulf of Mexico          下载免费PDF全文

Travis W. Washburn  Amanda W. J. Demopoulos  Paul A. Montagna 《Marine Ecology》2018,39(3)

There are thousands of seeps in the deep ocean worldwide; however, many questions remain about their contributions to global biodiversity and the surrounding deep‐sea environment. In addition to being globally distributed, seeps provide several benefits to humans such as unique habitats, organisms with novel genes, and carbon regulation. The purpose of this study is to determine whether there are unique seep macrobenthic assemblages, by comparing seep and nonseep environments, different seep habitats, and seeps at different depths and locations. Infaunal community composition, diversity, and abundance were examined between seep and nonseep background environments and among three seep habitats (i.e., microbial mats, tubeworms, and soft‐bottom seeps). Abundances were higher at seep sites compared to background areas. Abundance and diversity also differed among microbial mat, tubeworm, and soft‐bottom seep habitats. Although seeps contained different macrobenthic assemblages than nonseep areas, infaunal communities were also generally unique for each seep. Variability was 75% greater within communities near seeps compared to communities in background areas. Thus, high variability in community structure characterized seep communities rather than specific taxa. The lack of similarity among seep sites supports the idea that there are no specific infauna that can be used as indicators of seepage throughout the northern Gulf of Mexico, at least at higher taxonomic levels.  相似文献   

基于深度残差网络的机载LiDAR点云分类     

赵传  郭海涛  卢俊  余东行  张保明 《测绘学报》2020,49(2):202-213

机载LiDAR点云的分类是利用其进行城市场景三维重建的关键步骤之一。为充分利用现有的图像领域性能较好的深度学习网络模型,提高点云分类精度,并降低训练时间和对训练样本数量的要求,本文提出一种基于深度残差网络的机载LiDAR点云分类方法。首先提取归一化高程、表面变化率、强度和归一化植被指数4种具有较高区分度的点云低层次特征;然后通过设置不同的邻域大小和视角,利用所提出的点云特征图生成策略,得到多尺度和多视角点云特征图;再将点云特征图输入到预训练的深度残差网络,提取多尺度和多视角深层次特征;最后构建并训练神经网络分类器,利用训练的模型对待分类点云进行预测,经后处理得到分类结果。利用ISPRS三维语义标记竞赛的公开标准数据集进行试验,结果表明,本文方法可有效区分建筑物、地面、车辆等8类地物,分类结果的总体精度为87.1%,可为城市场景三维重建提供可靠的信息。  相似文献   

珊溪水库地区深部构造探测研究     

侯林锋  晏云翔  杨福平  余刚群  李林元 《大地测量与地球动力学》2018,38(8):818-822

通过对珊溪水库地区进行深地震反射探测，获得该地区近地表至地壳中部的精细几何结构和深浅构造关系。结果表明，该地区地壳结构在纵向具有明显的分带性，在横向显示出以断层为界的块状结构特点。研究区内走向北西的双溪-焦溪垟断层向深部延伸超过20 km进入中下地壳，为该区域内主要控制性断层，主要影响周围浅部断层的发育，是该地区主要发震构造。  相似文献   

It’s Time To Learn About Learning: Where Should the Environmental and Natural Resource Governance Field Go Next?     

Andrea K. Gerlak  Tanya Heikkila  Sharon L. Smolinski  Derek Armitage  Dave Huitema  Brendan Moore 《社会与自然资源》2019,32(9):1056-1064

This Commentary reflects on the state of the scholarship on learning for environmental and natural resource policy and governance. How have we been learning about learning? We highlight theoretical and empirical advancements related to learning, as well as areas of divergence between learning theories and frameworks, and underdeveloped knowledge around processes and outcomes. To address these limitations and improve progress in both theory and practice, we offer recommendations for learning scholarship by focusing on how to collectively engage in ‘learning about learning’.  相似文献   

Semantic segmentation of high spatial resolution images with deep neural networks     

Haiping Yang  Jiancheng Luo  Fang Chen 《地理信息系统科学与遥感》2019,56(5):749-768

Availability of reliable delineation of urban lands is fundamental to applications such as infrastructure management and urban planning. An accurate semantic segmentation approach can assign each pixel of remotely sensed imagery a reliable ground object class. In this paper, we propose an end-to-end deep learning architecture to perform the pixel-level understanding of high spatial resolution remote sensing images. Both local and global contextual information are considered. The local contexts are learned by the deep residual net, and the multi-scale global contexts are extracted by a pyramid pooling module. These contextual features are concatenated to predict labels for each pixel. In addition, multiple additional losses are proposed to enhance our deep learning network to optimize multi-level features from different resolution images simultaneously. Two public datasets, including Vaihingen and Potsdam datasets, are used to assess the performance of the proposed deep neural network. Comparison with the results from the published state-of-the-art algorithms demonstrates the effectiveness of our approach.  相似文献   

浮力装置触发深海管道水平向整体屈曲效果的数值模拟研究     

曹洋  张淑华  孙林 《海洋通报》2019,38(6):719-726

海洋油气资源的运输主要通过海底管道进行,管道在工作时受到较大的温度荷载,会产生整体屈曲变形。深海管道设计中常采用人为装置触发一定程度的水平向整体屈曲变形,来释放轴向的温度应力,浮力装置是常用的触发方式之一。本文通过数值模拟研究,分析了不同浮力大小和不同浮力施加范围下,管道水平向整体屈曲的临界屈曲力,得出临界屈曲力随浮力大小和施加范围变化的情况;并研究了不同土体阻力下,浮力装置触发整体屈曲的效果。研究表明,水平向土体阻力较大时,浮力装置触发水平向整体屈曲的效果较好。浮力装置的触发效果对轴向土体阻力不敏感。  相似文献   

Water mining from the deep critical zone by apple trees growing on loess     

Huijie Li  Bingcheng Si  Pute Wu  Jeffrey J. McDonnell 《水文研究》2019,33(2):320-327

There have been significant recent advances in understanding the ecohydrology of deep soil. However, the links between root development and water usage in the deep critical zone remains poorly understood. To clarify the interaction between water use and root development in deep soil, we investigated soil water and root profiles beyond maximum rooting depth in five apple orchards planted on farmland with stand ages of 8, 11, 15, 18, and 22 years in a subhumid region on the Chinese Loess Plateau. Apple trees rooted progressively deeper for water with increasing stand age and reached 23.2 ± 0.8 m for the 22‐year‐old trees. Soil water deficit in deep soil increased with tree age and was 1,530 ± 43 mm for a stand age of 22 years. Measured root deepening rate was far great than the reported pore water velocity, which demonstrated that trees are mining resident old water. The deficits are not replenished during the life‐span of the orchard, showing a one‐way mining of the critical zone water. The one‐way root water mining may have changed the fine root profile from an exponential pattern in the 8‐year‐old orchard to a relative uniform distribution in older orchards. Our findings enhance our understanding of water‐root interaction in deep soil and reveal the unintended consequences of critical zone dewatering during the lifespan of apple trees.  相似文献   

基于YOLOv3深度学习的海雾气象条件下海上船只实时检测     

王飞  刘梦婷  刘雪芹  秦志亮  马本俊  郑毅 《海洋科学》2020,44(8):197-204

海雾气象条件下船只高精度检测识别面临较大困难，传统的目标识别、定位方法效果差强人意。作者围绕海雾气象条件下不同类型船只的实时检测问题，提出一种基于YOLOv3深度学习的实时海上船只检测新思路。首先构建清晰图片和模糊图片（海雾、雨）的判别方法，实现图片清晰度分类处理；其次为提高海雾气象条件下海上船只的实时检测精度，消除海雾遮挡对目标识别的影响，运用暗通道先验去雾方法对含有海雾的图像实行去雾；最后基于YOLOv3深度学习算法对精细处理后的图像进行船只实时检测。实验结果表明该方法能够在海雾气象条件下高效、准确地检测到船只，对海上复杂环境条件下的船只实时检测研究具有一定的理论指导意义。  相似文献