建筑物立面分类与DIM点云滤波

针对目前基于近景摄影测量方法构建建筑物立面模型过程中因密集影像匹配（DIM）点云噪声所引起的建筑物立面TIN网格模型畸变问题,本文借鉴机器学习中样本学习的思想,对建筑物立面进行了分类并对DIM点云提出了相应的滤波方法,以达到去除DIM点云噪声和改善其TIN网格模型畸变的目的。其中,针对平面结构立面,采取先对点云样本进行学习计算构建数学立面模型所需参数,再对该立面模型设定阈值并对其点云进行滤波处理的方法;针对曲面结构立面,则结合DIM点云特性先将点云样本分类标记归为立面点与非立面点,再进行样本特征值学习,使用Logistic回归算法迭代计算求解最佳回归系数,从而构建滤波分类器的方法对立面点云进行滤波处理。试验结果表明,本文滤波处理方法能将立面DIM点云噪声有效识别并去除,而且使用该方法处理后所得点云构建的建筑物立面TIN网格模型精细化程度得到有效提高,模型质量得到明显改善。     

基于卷积神经网络的无人机影像违章建筑检测应用

无人机航拍影像具有分辨率高、回访周期短等特点,利用无人机遥感技术手段对城市范围的建设进行动态监测,可及时、有效地发现涉嫌违法的建设活动。本文结合实际项目需求,研究通过卷积神经网络方法进行违章建筑的自动检测,替代过去靠大量人力检查的模式,目前测试区域无人机影像试验取得了较好的效果,在样本数据不足5000份的情况下,准确率和召回率分别达到了71%和88%。随着样本数据的不断增多,基于该深度学习方法将较大程度上持续提升检测准确率和召回率,能够更精准地发现违法活动,具有较大的实际应用价值及潜力。     

深度学习PaddlePaddle框架支持下的遥感智能视觉平台研究与实现

百度深度学习PaddlePaddle框架支持下的遥感智能视觉平台,能够运用深度学习技术实现遥感影像的智能建模、训练和解译。本文通过深入分析PaddlePaddle图像分割模型库PaddleSeg的图像处理深度学习算法模型DeepLabV3+、U²-Net及RetinaNet,开发设计了遥感智能视觉平台,实现了遥感影像的地块分割、变化检测和斜框检测等专业功能。研究表明：遥感智能视觉平台提取的图斑总面积是目视解译的80%、有效图斑比例为76%、错误图斑比例为18%,实现了快速有效的遥感图像智能处理。     

深度学习支持下的丘陵山区耕地高分辨率遥感信息分区分层提取方法

耕地是丘陵山区稀缺的土地资源,具有地形条件复杂、种植结构多样的特点,导致了山地耕地信息难以快速、准确获取,并且基于传统的遥感数据及遥感监测方法开展山区耕地信息快速自动提取比较困难。针对这一问题,本文以西南山区贵州省息烽县作为试验区,根据地理空间异质性特征,提出分区控制、分层提取的耕地形态信息提取思路,构建了一种地貌单元约束条件下的分区分层耕地形态信息的提取方法。该方法首先根据地貌-植被特征将试验区划分为平坝区、山坡区、林草区3类地理分区;然后在每类分区基础上,根据耕地所呈现的视觉特征划分为不同的类型,对不同类型的耕地分别设计不同的深度学习模型进行分层提取。试验结果证明,该方法对山区复杂地形背景噪声具有较好的抑制作用,所提取的耕地地块信息相比于传统方法更符合实际耕地的实际分布形态,有效地减少了漏提率和错提率。     

数据驱动的蒸散发遥感反演方法及产品研究进展

蒸散发是水圈、大气圈和生物圈中水分循环和能量交换的纽带。在全球尺度上,蒸散发约占陆地降水总量的60%;作为其能量表达形式,潜热通量约占地表净辐射的80%。随着通量观测技术的发展,全球长期持续的观测数据得以获取和共享,近年来基于数据驱动的蒸散发遥感反演方法取得了较好的研究进展。本文针对数据驱动的蒸散发遥感反演方法和产品,从经验回归、机器学习和数据融合3个方面展开,对现有的研究进展进行了梳理、归纳和总结,并从驱动数据、反演方法、已有产品等方面指出目前仍存在的问题和不足。未来仍需开展数据驱动的高时空分辨率的蒸散发遥感反演方法的研究,有效考虑地表温度和土壤水分等可以指示地表蒸散发短期变化的重要信息,同时加强基于过程驱动的物理模型与数据驱动的模型的结合,使两类模型能互为补充、各自发挥所长,共同推动蒸散发遥感反演研究水平的进步。     

Data selection using support vector regression

Geophysical data sets are growing at an ever-increasing rate, requiring computationally efficient data selection(thinning)methods to preserve essential information. Satellites, such as Wind Sat, provide large data sets for assessing the accuracy and computational efficiency of data selection techniques. A new data thinning technique, based on support vector regression(SVR), is developed and tested. To manage large on-line satellite data streams, observations from Wind Sat are formed into subsets by Voronoi tessellation and then each is thinned by SVR(TSVR). Three experiments are performed. The first confirms the viability of TSVR for a relatively small sample, comparing it to several commonly used data thinning methods(random selection, averaging and Barnes filtering), producing a 10% thinning rate(90% data reduction), low mean absolute errors(MAE) and large correlations with the original data. A second experiment, using a larger dataset, shows TSVR retrievals with MAE < 1 m s-1and correlations 0.98. TSVR was an order of magnitude faster than the commonly used thinning methods. A third experiment applies a two-stage pipeline to TSVR, to accommodate online data. The pipeline subsets reconstruct the wind field with the same accuracy as the second experiment, is an order of magnitude faster than the nonpipeline TSVR. Therefore, pipeline TSVR is two orders of magnitude faster than commonly used thinning methods that ingest the entire data set. This study demonstrates that TSVR pipeline thinning is an accurate and computationally efficient alternative to commonly used data selection techniques.     

The Coral Triangle Initiative and regional exchanges: Strengthening capacity through a regional learning network

Natural resource management and conservation programs that promote building capacity and social learning among participants often lead to the formation of learning networks: a type of social network where learning is both a goal and potential outcome of the network. Through forming relationships and sharing information, participants in a learning network build social capital that can help a network achieve social and environmental goals. In this study, we explored social capital in a learning network that emerged through a large-scale marine governance effort, the Coral Triangle Initiative on Coral Reefs, Fisheries, and Food Security. Through a mixture of social network analysis and key informant interviews, we examined the major patterns of information exchange among individuals who had participated in regional learning exchanges; evaluated whether the network's structure resulted in information sharing; and considered implications for strengthening network sustainability, capacity building, and learning. We found that the Regional Exchange network fostered information sharing among participants across national and organizational boundaries. While the network had individuals who were more central to information sharing, the network structure was generally decentralized, indicating potential resilience to changes in leadership and membership. Participants stressed the importance of the knowledge and connections they had acquired through the learning network; however, they expressed doubts regarding its sustainability and stressed the need for a strong coordinating entity. Our findings suggest that conservation learning networks have the ability to bridge cultural divides and promote social learning; however, a strong network coordinator and continuing efforts to support information sharing and learning are crucial to the network's strength and sustainability. The tangible learning and capacity development outcomes cultivated through Regional Exchange network underscore the value of and need to invest in conservation networks that support peer-to-peer learning.     

The PISA grammar decodes diverse human–environment approaches

Human–environment interactions are studied by several groups of scholars who have elaborated different approaches to describe, analyze, and explain these interactions, and eventually propose paths for management. The SETER project (Socio-Ecological Theories and Empirical Research) analyzed and compared how “flag-holders” of distinct school of thought in human–environment scholarship approached a number of empirical problems of environmental management. This paper presents the findings from this experiment by concentrating on how representatives of four schools of thought approached one of these case studies: the plant health crisis in greenhouse tomato production in south of France. Our analysis suggests that these approaches share a common conceptual vocabulary composed of four explanatory elements of change (Power, Incentives, System and Adaptation-PISA). We argue that what distinguishes these schools from one another is the syntax—the “rules” by which researchers in each of the sub-disciplines tend to organize the components of this shared conceptual vocabulary. In other words, the schools under scrutiny are differentiated not so much by what they speak of, but rather in what order, or hierarchy, do they tend to rank the importance and/or the sequence of each of these concepts in human–environment explanations. The results of our experiment support the view that communication and cooperation across the diverse human–environment traditions is possible and productive. At the same time, however, we argue that it is the distinctiveness of the claims yielded by these different schools of thought that augment our collective understanding of complex socio-ecological problems. Attempts to integrate these perspectives in one unitary approach would undermine the intellectual wealth necessary to meet the challenges of the Anthropocene.     

Local collective action: Adaptation to coastal erosion in the Monkey River Village,Belize

Successful adaptation to environmental change and variability is closely connected with social groups’ ability to act collectively, but many social-ecological challenges exceed local adaptive capacity which necessitate assistance from governmental institutions. Few studies have investigated how local collective action can be used to enrol external support for adaptation. This paper reduces this research gap by analysing a locally driven adaptation process in response to coastal erosion in Monkey River Village, Belize. Drawing on literature on adaptation and political ecology, we examine the different strategies the local residents have used over time to influence government authorities to support them in curbing the coastal erosion. Our findings show that the local mobilisation generated government support for a temporary sea defence and that collective strategies emerge as a response to threats to a place specific way of life. Our case illustrates that it was essential that the villagers could ally with journalists, researchers and local NGOs to make their claims for protection heard by the government. The paper contributes to adaptation research by arguing that local collective action, seen as contestation over rights to protection from environmental change, can be a means for places and communities not prioritised by formal policies to enrol external support for adaptation. Our study supports and adds to the perspective that attention to formal arrangements such as adaptation policy alone has limited explanatory power to understand collective responses to change.     

基于深度学习的循环水养殖鳗鲡(Anguilla)计数研究

鳗鲡(Anguilla)作为我国优质水产养殖种类,精准掌握其数量对高效养殖有重要意义。为实现对循环水养殖鳗鲡的准确计数,提出了一种基于深度学习的改进Faster RCNN模型。针对检测目标即鳗鲡头部尺寸小的问题,选择在特征提取网络ResNet50中加入FPN结构来作为模型的骨干网络,以提取并融合多尺度的特征;针对原模型锚框都是基于人工经验设置的,并不适用于鳗鲡数据集的问题,使用k-means聚类算法对训练集中标注的鳗鲡头部检测框进行聚类分析,获得了适合鳗鲡数据集的15种不同尺度的锚框;针对图像中存在鳗鲡头部重叠的问题,选择使用Soft-NMS算法替代原NMS算法对RPN部分生成的候选框进行筛选,以减少模型对鳗鲡重叠部分的漏检情况。试验结果表明:改进后的Faster RCNN模型对鳗鲡头部的检测精度(mAP^0.5)高达96.5%,较原Faster RCNN模型(Backbone为ResNet50)显著提升了14%,与SSD300和YOLOV3模型相比分别显著提升了24.9%和15%;在鳗鲡计数上,利用改进后的Faster RCNN模型检测结果进行计数,计数准确率达到90%以上,提升了模型对鳗鲡的检测识别能力。