Comprehensive evaluation on urban sustainable development of Harbin City in Northeast China

It is an effective way in realizing urban coordinated and sustainable development to establish a series of indicators and to evaluate urban environmental and socioeconomic development. According to the characteristics of Harbin City in Northeast China, an indicator system including five subsystems and 37 indicators was established for comprehensive evaluation on urban sustainable development. The development indexes of all urban subsystems and complex system were calculated quantitatively using the comprehensively integrated methods composed of Principle Component Analysis, Analytic Hierarchy Process and weighed index method, and then the comprehensive level of urban sustainable development and the degree of urban interior coordination were analyzed. The results indicated that 1) the overall urban development presented an uptrend, however, the interior development was not well balanced from 1996 to 2006; 2) the development in each subsystem presented a strong fluctuation; and 3) the development in resources subsystem showed a downtrend. Based on those results, the suggestions of urban sustainable development were put forward at the end.     

基于机器学习中分辨率遥感影像分类应用研究

地表信息提取是地理国情监测和环境保护的重要环节,本文选用Landsat-8 OLI遥感影像为主要数据源,针对分类样本数据少这一问题,在对2015年和2017年全球地表覆盖数据优化处理的基础上,将其作为分类的先验知识,以最大似然法、支持向量机(Support Vector Machine,SVM)和随机森林(Random Forest,RF)机器学习方法,以及基于VggNet-16和ResNet-18模型的深度学习方法,对黄河三角洲区域进行地表信息提取方法的应用研究.结果表明,在利用粒子群优化遗传算法进行特征优化的基础上,RF和SVM模型分类总体精度较高,其中RF和SVM模型分类总体精度最高可达87.3％、86％;VggNet-16和ResNet-18两种模型的分类精度较机器学习方法有较大提升,其中以ResNet-18分类精度最佳,最高总体精度可达94.1％,Kappa系数为0.91.本文的研究方法在中分辨率遥感地物分类时具有较好的适用性和推广价值.     

太湖水质参数MODIS的遥感定量提取方法

本文利用MODIS数据的可见光、近红外波段和准实时的地面采样数据,分别利用线性回归模型和神经网络模型反演了太湖的叶绿素a和悬浮物浓度.结果表明,利用MODIS数据的波段组合(M2/M8)和(M2/M9)可估算太湖的叶绿素a浓度;而MODIS数据的波段组合(M12/M17)、(M13/M17)及MODIS(M4)波段能定量估算太湖的悬浮物浓度,但估算精度仍不能满足实际需要.因此,构建了一个以MODIS可见光及近红外波段为输入,以太湖水质参数为输出的2层BP神经网络模型反演太湖的水质参数,大大提高了反演精度.     

会“开花”的石头 会找矿的“花朵”

岩石也会"开出"艳丽的花朵.当岩石中有向三度宅问放射生长的矿物品体时,无论把岩石从什么方向剖开,只要断面经过矿物晶体,都会在岩石断面上出现花朵般的图案.如果矿物品体呈柱状, "花瓣"则又细又长,图案好似菊花;如果矿物晶体呈片状, "花瓣"则短而胖,"花朵"就如同攻瑰花.     

Mapping the farming-pastoral ecotones in China

Conventional farming-pastoral ecotones methods of delineating were not quantitative and could not fully show their spatial distribution. The present paper attempts to develop quantitative methods for mapping farming-pastoral ecotones in China. Nine indicators, related to temperature, precipitation and altitude aspects, were selected to quantify ecological susceptibility of vegetation （crops and forage）. Methods of analytic hierarchy process （AHP） and expert score ranking combined with fuzzy set theory were applied to assign the weight for each indicator and to define the membership functions. The geographic information system （GIS） was used to manage the spatial database and conduct the spatial analysis. According to the spatial calculation of evaluation model integrated with GIS, the ecological susceptibility of vegetation （crops and forage） was mapped. Three different zones, pastoral area, farming-pastoral ecotones and farming area, were classified by spatial cluster analysis and the maximum likelihood classification for the numeric map of vegetation ecological susceptibility by GIS. This map was validated by the economic statistical result based on the ratio of the output value from animal husbandry in total output value of agriculture by the National Bureau of Statistics in China, indicating that the mapping of the farming-pastoral ecotones may be accepted.     

加权损失函数在划分甘蔗种植区域的应用——以广西扶绥县某甘蔗种植区域为例

文章针对深度学习中样本数量不均衡导致训练出来的神经网络模型(以下简称模型)性能下降、精度不高等问题,根据不同占比的样本数量设置不同的权重值,通过权重来控制不同类别的损失,调整原有交叉熵损失函数的计算结果,将训练权重更多地向样本数量较少的一类倾斜,从而使训练过程趋于平衡,进而不断提高模型的精度指标.此次实验选择了广西扶绥...     

基于光谱特征的湿地湿生植物信息提取研究

基于湿生植物光谱特征分析,采用面向对象分类方法提取湿地典型植被信息。在大量的野外实地调查基础上,确定提取的典型植物类型为芦苇(Phragmites australis)、长芒稗(Echinochloa crusgallii)和睡莲(Nymphaeatetragona);采集湿地优势植物光谱数据,将优势植物与提取对象芦苇、长芒稗和睡莲的光谱特征进行相关性分析,获取物种间区分性好的波段及波段组合,参与影像分割权重的设置;根据典型植被的分布特征,确定面向对象分割尺度(其中芦苇的分割尺度为50,长芒稗的分割尺度为20,睡莲的分割尺度为100)。通过研究发现:基于光谱特征分析的面向对象的分类精度为96%,而未利用光谱特征的面向对象的分类精度为87.3%,传统监督分类精度仅为82.3%。证明在面向对象提取前对植物光谱特征分析得到区分性好的波段及波段组合参与分割,对提高分类精度起关键作用。     

单频机载激光测深海陆回波自动分类方法EI北大核心CSCD

海陆回波分类是机载激光测深中的一项波形预处理步骤,关系着后续信号检测和点云生成的精度。针对现有海陆回波分类方法不适用于单频机载激光测深系统且自动化程度不高的问题,本文提出一种单频机载激光测深海陆回波自动分类方法:首先,通过首末回波信号检测及点位计算获得回波的点云高程特征;然后,采用高程直方图拟合的方式确定平均水面位置,依据点云高程特征判定大部分回波的海陆属性,对余下的未定回波,仅保留其中的最强信号并统一处理为单信号回波,同时提取波形的信号特征和能量分布特征,依据点云高程特征的相似性自动建立训练样本集;最后,利用支持向量机分类器实现未定回波的分类。采用国产系统Mapper5000采集的实测数据进行试验,结果表明基于首末回波点云的初分类可快速、准确地对远离海陆交界处的回波进行分类,基于波形特征的未定回波分类可在自动建立的训练样本集支持下实现海陆交界处未定回波的高精度分类。与传统方法相比,本文方法无须近红外通道波形和人工样本的辅助就可以达到较高的分类精度,其中总体分类精度可达99.82%,海陆交界处分类精度可达91.59%。     

海拉尔盆地上库力组重矿物组合特征

海拉尔盆地兴安岭群上库力组自下而上可分出5个重矿物组合：磁黄铁矿-锆石-白钛石-锐钛矿组合,黄铁矿高含量组合,黑云母-白钛石-锆石-钛铁矿组合,菱铁矿-白钛石-黑云母-电气石组合,白钛石-黑云母-菱铁矿-石榴子石组合.不稳定矿物含量高,不同层段呈现高低韵律性变化则与火山-沉积作用强弱交替变化的环境有关.较高的ATi指数反映母岩应为中-酸性火山岩.ZTR指数不高说明重矿物的成熟度较低.由下至上陆源重矿物变少,尤其是上库力组D段（凝灰岩段）,自生黄铁矿占很高的百分比.