基于决策树特征提取的支持向量机在岩性分类中的应用

由于支持向量机属于黑箱模型,因此在进行模型学习时无法直接对特征进行选择,而决策树模型在递归创建的过程中自身具有一定的特征选择能力。针对岩性分类问题,本文将决策树和支持向量机结合,通过决策树的建立,在考虑特征重要性的前提下,利用树节点的高度对特征进行提取,并将具有更高分类能力的特征送入支持向量机进行岩性分类。结果表明:通过决策树的特征提取,减少了支持向量机模型的输入特征,从而有效控制了模型的复杂度,使得模型更加稳定并具有更高的分类精度,测试集精度能够提升10%以上。     

地图数据库中的结构化河网及其自动建立

本文提出了一种基于河段的结构化河网,讨论了这种结构化河网的自动建立过程,主要包括两个结构索引即线索树结构索引及层次结构索引的自动产生;根据包含以上两个索引的结构化河网,提供了若干检索函数,作为地图数据库中结构处理时结构信息的提取手段。     

轨道扣件病害自动化检测研究及应用

传统扣件检查基本采用人工检查方式,不仅工作效率低、劳动强度大,而且人为干扰因素多、检查采样率低.针对以上问题,本文提出了一种基于线结构光传感器的轨道扣件损伤和松动检测方法,并开发了轨道扣件智能监测系统以实现扣件病害自动检测.最终将研究成果应用于徐州市某地铁区间轨道扣件检测,验证了该方法的可行性和准确率,为地铁轨道扣件检...     

基于梯度提升算法的岩性识别方法

传统的岩性识别方法如岩屑录井、钻井取心及测井资料解释等技术,对录井质量的依赖程度较高,识别精度与效率低,泛化能力差。随着计算机技术的迅速发展,将测井资料与计算机技术相结合开展岩性研究已成为岩性识别的有效手段。本文提出了一种基于梯度提升算法XGBoost和LightGBM的岩性识别方法。以苏里格气田苏东41-33区块下碳酸盐岩储层为例进行测试验证,采用该方法结合测井资料中的声波时差、自然伽马、光电吸收截面指数、密度、深侧向电阻率和补偿中子等6种参数进行岩性识别,并与KNN （K近邻分类器）、朴素贝叶斯和支持向量机等传统算法进行对比,结果表明,3种传统算法的岩性识别准确率分别为78.45%、74.43%和78.72%,基于梯度提升算法XGBoost和LightGBM的识别准确率分别达到了98.90%和98.72%,远高于传统算法。     

A study on the bio-hydro mechanical effects of the tree roots on the Agaclı coal reclaimed – forested land

The research area that is in the north of Istanbul (Turkey) and it is a land of open pit coal mining residuals reclaimed and turned to forest in 1988–1989. The materials that are open pit mine spoils are formed by sandy loam, sandy clay loam, heavy loam and clay (noncalcareous Pliocene I sediments). Pseudoacacia (Robinia pseudoacacia), Maritima Pine (Pinus pinaster) and Stone Pine (Pinus pinea) are planted on these materials. The aims of this research were to determine the strength of the material provided by tree roots and evaluate the performance of the forestation practices at lands especially where open pit coal mining spoil (residual) materials piled up. The research has been performed at two stages. The first one was hydro-mechanical effect concerned about material moisture depletion as a result of transpiration and the second one was bio-mechanical effects due to strength parameters of the roots and their growth pattern within the material. All results were evaluated in relation to tensile strength and displacement curves due to pre and post failure behaviors of materials and root systems. The results showed the increase of the displacement of the material was the significant indicator of the bio-mechanics tensile strength of the material provided by the tree roots. The fast growing type of tree roots have stabilized the material up to a depth for 80–100 cm by covering the coal residual materials within 12 years.     

基于MODIS影像多特征的CART决策树分类

以山东省为研究区域,利用2009年9月MODIS的8 d合成波段反射率产品MOD09,选择特征变量植被指数(NDVI、EVI)、NDWI、NDMI、NDSI及辅助信息DEM,通过选取其中的影像特征组合来确定分类方案,构建各波段组合的CART决策树,对MODIS影像进行分类,得到CART决策树的最优波段组合。结果表明,特征变量DEM、NDVI、EVI对分类结果贡献较大;将CART决策树的分类结果与其相对应的最大似然分类结果进行比较可知,基于影像多特征的CART决策树分类方法能明显提高分类精度。     

利用多源遥感数据进行城市不透水面覆盖度估算

利用深圳实验区4种不同传感器获取的遥感数据,通过CART算法进行城市ISP估算。讨论了多光谱遥感数据的不同波段在ISP估算中的重要性,比较了针对三种不同中分辨率影像建立的ISP估算模型在性能上的差异。实验结果表明,近红外波段对ISP估算结果的贡献最大,具有较高空间分辨率和成像辐射质量的遥感影像得到的估算结果精度较高,所有的估算结果均在实际ISP分布范围的两端分别存在着高估和低估的现象。     

机载激光雷达平均树高提取研究

为了研究机载激光雷达(LiDAR)树高提取技术,以山东省泰安市徂徕山林场为实验区,于2005年5月进行了机载LiDAR数据获取和外业测量.通过对LiDAR点云数据的分类处理,分别得到了试验区的地面点云子集、植被点云子集和高程归一化的植被点云子集.基于高程归一化的植被点云子集计算了上四分位数处的高度,与实地测量的数据进行了比较,并结合中国森林调查规程进行了实用性分析.结果表明:对于较低密度的点云数据,使用分位数法可以较好地进行林分平均高的估计;机载激光雷达技术对树高估计是可行的,精度都高于87%,总体平均精度为90.59%,其中阔叶树的精度高于针叶树.该试验精度可以满足中国二类森林调查规程中平均树高因子的一般商品林和生态公益林的精度要求,对国有商品林小班的调查精度要求(5%)存在一点差距,需要在国有商品林区进一步开展验证工作.对本试验区而言,已经可以满足其作为森林公园生态公益林的调查要求.     

结合地形因子与分层策略的喀斯特地区无人机影像分类

西南喀斯特山区地形起伏较大,地物分布较为破碎,致使传统的光谱特征一次分类方法的精度较低。本文基于高分辨率无人机正射影像和地形指标,充分利用无人机遥感影像空间特征、光谱特征、纹理特征及地形特征,采取面向对象CART决策树算法与分层策略提取了研究区土地覆盖类型。研究表明,结合空间地形因子和分层策略的方法减少了破碎区地物间的相干扰,故具有较高的分类精度,总体分类精度达91.2%,Kappa系数为0.87,较传统一次分类精度提高了9.8%,Kappa系数提高了0.13。该方法对西南喀斯特地区土地覆盖解译精度较好,可为土地利用监测提供参考。     

Microcoleus autumnalis and filamentous algae-dominated mats and chlorophyll-a increase with agricultural land use but respond differently to associated nutrient and sediment enrichment

ABSTRACT

This study investigated the impact of pastoral land use and nutrient and fine sediment inputs on Microcoleus autumnalis and filamentous algae-dominated mats, and benthic chlorophyll-a in streams (lower North Island, New Zealand). Surveying and sampling was undertaken monthly at 61 sites spanning a wide gradient in catchment cover and environmental conditions. Two boosted regression tree models were built. The first models included pastoral land cover and five environmental variables as predictors. In the second model pastoral land cover was replaced by nutrient/sediment data. The abundance of the two mat types and chlorophyll-a increased when pastoral land cover was between 20% and 70% (model 1). Replacement of pastoral land cover by nutrient/sediment data (model 2) slightly improved the model fit for all three periphyton variables. Microcoleus autumnalis-dominant mats increased with dissolved inorganic nitrogen concentrations up to ca. 0.6?mg?L^?1, and in streams with more frequent flushes. In contrast, filamentous algal-dominated mats increased with turbidity, and in streams with less frequent flushes. Chlorophyll-a generally followed the response of the dominant periphyton type. Increased knowledge on responses of specific periphyton types, rather than total biomass, to environmental variables is essential to guide effective management strategies.