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   

Performance of small-scale irrigation schemes in Lake Tana Basin of Ethiopia: technical and socio-political attributes     

Abebech Abera  Niko E.C. Verhoest  Seifu A. Tilahun  Tena Alamirew  Enyew Adgo  Michael M. Moges 《自然地理学》2019,40(3):227-251

This study was conducted to investigate technical and socio-political attributes that lead to the underperformance of two selected irrigation schemes (Shina and Bebeks) in the Lake Tana floodplains, Ethiopia. Irrigation application efficiency (AE) at nine experimental fields showed a wide range, from 20 to 80%, but was mostly between 40 and 60%. Irrigation water-use efficiency (IWUE) varied from 1.9 to 7.2 kg m^?3 for onion and 0.9 to 1.2 kg m^?3 for maize. The lined and earthen canal conveyance losses in Bebeks were 0.037 and 0.047 l s^?1 m^?1, whereas in Shina they were 0.033 and 0.044 l s^?1 m^?1, respectively. The overall consumed ratio (OCR) of water was 0.58 for Bebeks and varied from 0.73 to 1.2 in Shina. Both schemes are performing below the standard based on technical performance indicators. Irrigation water user associations (WUAs) were not implemented, but irrigation committees (ICs), composed of local political leaders, are managing both schemes. Canal and reservoir sedimentation from erosion of upstream catchment areas during the rainy season was the major problem.  相似文献   

中国城市体系网络化研究   总被引：14，自引：1，他引：13  

胡国建  陈传明  金星星  王强 《地理学报》2019,74(4):681-693

当前,学者普遍认同城市体系研究的重心从等级范式向网络范式转向。在大量的文献梳理基础上,阐述20世纪80年代以来中国城市体系研究范式的转向过程,并进一步梳理中国城市网络研究的进展、问题与关注重点。结果表明：2010年之后,网络成为中国城市体系研究的主流范式,等级体系文献数量呈波动递减趋势,全国和长三角成为热点关注区域;属性数据一定程度上弥补了早期城市网络研究中数据匮乏的状况,关系数据应成为未来研究的侧重点;不同研究视角及其所应用的数据存在着一定的局限性,多元视角的综合研究成为趋势。在此基础上,提出城市网络研究未来可能的突破点,包括：调整国家城市网络的研究框架、超越交通数据直接考察城市间的实体联系、测度企业外部的横向联系、有向加权网络分析和可视化、增加研究中的地理“意味”等。  相似文献   

SAR image despeckling with a multilayer perceptron neural network     

Xiao Tang  Lei Zhang  Xiaoli Ding 《International Journal of Digital Earth》2019,12(3):354-374

Speckle noise in synthetic-aperture radar (SAR) images severely hinders remote sensing applications; therefore, the appropriate removal of speckle noise is crucial. This paper elaborates on the multilayer perceptron (MLP) neural-network model for SAR image despeckling by using a time series of SAR images. Unlike other filtering methods that use only a single radar intensity image to derive their parameters and filter that single image, this method can be trained using archived images over an area of interest to self-learn the intensity characteristics of image patches and then adaptively determine the weights and thresholds by using a neural network for image despeckling. Several hidden layers are designed for feedforward network training, and back-propagation stochastic gradient descent is adopted to reduce the error between the target output and neural-network output. The parameters in the network are automatically updated in the training process. The greatest advantage of MLP is that once the despeckling parameters are determined, they can be used to process not only new images in the same area but also images in completely different locations. Tests with images from TerraSAR-X in selected areas indicated that MLP shows satisfactory performance with respect to noise reduction and edge preservation. The overall image quality obtained using MLP was markedly higher than that obtained using numerous other filters. In comparison with other recently developed filters, this method yields a slightly higher image quality, and it demonstrates the powerful capabilities of computer learning using SAR images, which indicate the promising prospect of applying MLP to SAR image despeckling.  相似文献   

基于遗传算法优化BP神经网络的玉米遥感估产方法     

于海洋  陈圣波  杨北萍  安秦 《世界地质》2020,39(1):208-214

以内蒙古自治区开鲁县玉米作物为研究对象,将生育期内玉米遥感影像所提取的多种植被指数和实地采样点的测产数据作为训练值,利用BP(back propagation)神经网络和遗传算法优化BP(GA-BP)神经网络估产模型,得出网络预测的玉米产量数值。通过决定系数R 2和均方根误差RMSE,比较实测产量与预测产量之间的精度,BP神经网络模型R^2为0.8452,RMSE(%)为28.37;遗传算法优化BP神经网络模型R^2为0.9850,RMSE(%)为6.70,表明遗传算法优化BP神经网络估产模型具有一定可行性和可信度。  相似文献   

Dynamic,discontinuous stream networks: hydrologically driven variations in active drainage density,flowing channels and stream order          下载免费PDF全文

S. E. Godsey  J. W. Kirchner 《水文研究》2014,28(23):5791-5803

Despite decades of research on the ecological consequences of stream network expansion, contraction and fragmentation, surprisingly little is known about the hydrological mechanisms that shape these processes. Here, we present field surveys of the active drainage networks of four California headwater streams (4–27 km²) spanning diverse topographic, geologic and climatic settings. We show that these stream networks dynamically expand, contract, disconnect and reconnect across all the sites we studied. Stream networks at all four sites contract and disconnect during seasonal flow recessions, with their total active network length, and thus their active drainage densities, decreasing by factors of two to three across the range of flows captured in our field surveys. The total flowing lengths of the active stream networks are approximate power‐law functions of unit discharge, with scaling exponents averaging 0.27 ± 0.04 (range: 0.18–0.40). The number of points where surface flow originates obey similar power‐law relationships, as do the lengths and origination points of flowing networks that are continuously connected to the outlet, with scaling exponents averaging 0.36–0.48. Even stream order shifts seasonally by up to two Strahler orders in our study catchments. Broadly, similar stream length scaling has been observed in catchments spanning widely varying geologic, topographic and climatic settings and spanning more than two orders of magnitude in size, suggesting that network extension/contraction is a general phenomenon that may have a general explanation. Points of emergence or disappearance of surface flow represent the balance between subsurface transmissivity in the hyporheic zone and the delivery of water from upstream. Thus the dynamics of stream network expansion and contraction, and connection and disconnection, may offer important clues to the spatial structure of the hyporheic zone, and to patterns and processes of runoff generation. Copyright © 2014 John Wiley & Sons, Ltd.  相似文献   

水文监测网络运行状态综合评估系统设计与实现     

王菊  胡超  金惠英  梁妙元  李胜 《水文》2021,41(1):55-60

对水文监测网络的实时状态进行综合评估,从而掌握网络的整体性能是对水文监测网络进行管理的重要方面。针对当前水文监测网络综合评估缺少合理有效评价机制的现状,设计并开发水文监测网络实时状态综合评估系统,实现对水文网络中测站传感器、RTU、信道状态的采集、评估和综合呈现,提高对水文监测网络的分析水平。  相似文献   

Late Quaternary chironomid community structure shaped by rate and magnitude of climate change     

Roseanna J. Mayfield  Peter G. Langdon  C. Patrick Doncaster  John A. Dearing  Rong Wang  Gaute Velle  Kimberley L. Davies  Stephen J. Brooks 《第四纪科学杂志》2021,36(3):360-376

Much is known about how climate change impacts ecosystem richness and turnover, but we have less understanding of its influence on ecosystem structures. Here, we use ecological metrics (beta diversity, compositional disorder and network skewness) to quantify the community structural responses of temperature-sensitive chironomids (Diptera: Chironomidae) during the Late Glacial (14 700–11 700 cal a bp ) and Holocene (11 700 cal a bp to present). Analyses demonstrate high turnover (beta diversity) of chironomid composition across both epochs; however, structural metrics stayed relatively intact. Compositional disorder and skewness show greatest structural change in the Younger Dryas, following the rapid, high-magnitude climate change at the Bølling–Allerød to Younger Dryas transition. There were fewer climate-related structural changes across the early to mid–late Holocene, where climate change was more gradual and lower in magnitude. The reduced impact on structural metrics could be due to greater functional resilience provided by the wider chironomid community, or to the replacement of same functional-type taxa in the network structure. These results provide insight into how future rapid climate change may alter chironomid communities and could suggest that while turnover may remain high under a rapidly warming climate, community structural dynamics retain some resilience.  相似文献   

超参数对GRU-CNN混合深度学习弹性阻抗反演影响研究     

梁立锋  刘秀娟  张宏兵  陈程浩  陈锦华 《物探与化探》2021,(1):133-139

CNN-GRU混合深度学习反演弹性阻抗取得了较好的反演效果。但是,基于深度学习的叠前反演参数众多,包括内部深度学习网络可学习参数和外部超参数等,目前超参数选取对网络性能及计算速度影响尚缺乏系统性研究,这直接影响到了该方法的进一步推广应用。因此,本文在混合深度学习反演弹性阻抗基础上,探讨学习率、Epoch、batch_size、正则化参数及参与网络训练的测井个数等5个超参数对网络性能及计算速度的影响,为深度学习地震反演超参数选取提供依据。研究结果可为三维大面积深度学习反演提供一个可行的质控手段,对于推动深度学习方法在石油物探中广泛应用具有一定意义。  相似文献   

Distribution Prediction of Middle-deep Lacustrine Source Rocks in Eocene Wenchang Formation in the Kaiping Sag,Pearl Mouth Basin     

XIONG Lianqiao  XIE Xiaojun  ZHAO Zhao  ZHANG Chunyu  WANG Long  LIAO Jihu  CAI Lulu  XU Wei 《《地质学报》英文版》2021,95(1):248-258

Oil and gas shows are rich in drilling wells in Kaiping sag,however,large oilfield was still not found in this area.For a long time,it is thought that source rocks were developed in the middle-deep lacustrine facies in the Eocene Wenchang Formation,while there is no source rocks that in middle-deep lacustrine facies have been found in well.Thickness of Wenchang Formation is big and reservoirs with good properties could be found in this formation.Distribution and scale of source rock are significant for further direction of petroleum exploration.Distribution characterization of middle-deep lacustrine facies is the base for source rock research.Based on the sedimentary background,fault activity rate,seismic response features,and seismic attributes were analyzed.No limited classification method and multi-attributes neural network deep learning method were used for predicting of source rock distribution in Wenchang Formation.It is found that during the deposition of lower Wenchang Formation,activity rate of main fault controlling the sub sag sedimentation was bigger than 100 m/Ma,which formed development background for middle-deep lacustrine facies.Compared with the seismic response of middle-deep lacustrine source rocks developed in Zhu I depression,those in Kaiping sag are characterized in low frequency and good continuity.Through RGB frequency decomposition,areas with low frequency are main distribution parts for middle-deep lacustrine facies.Dominant frequency,instantaneous frequency,and coherency attributes of seismic could be used in no limited classification method for further identification of middle-deep lacustrine facies.Based on the limitation of geology knowledge,multi-attributes of seismic were analyzed through neural network deep learning method.Distribution of middle-deep lacustrine facies in the fourth member of Wenchang Formation is oriented from west to east and is the largest.Square of the middle-deep lacustrine facies in that member is 154 km²and the volume is 50 km³.Achievements could be bases for hydrocarbon accumulation study and for exploration target optimization in Kaiping sag.  相似文献