B-PNN改进算法在土地估价信息系统的应用

人工神经网络是解决主观因素影响过大和智能化程度不高这两个问题的合适方法。主要采用人工神经网络中的B-P神经网络建立土地估价智能模型,在ArcGIS的开发平台下,实现了基于B-P神经网络改进算法的人工神经网络土地估价模型。研究得出,利用人工神经网络可以解决土地估价过程中主观因素影响过多这一问题,可以提高城镇土地定级估价信息系统智能化程度。     

Single tree biomass modelling using airborne laser scanning

Accurate forest biomass mapping methods would provide the means for e.g. detecting bioenergy potential, biofuel and forest-bound carbon. The demand for practical biomass mapping methods at all forest levels is growing worldwide, and viable options are being developed. Airborne laser scanning (ALS) is a promising forest biomass mapping technique, due to its capability of measuring the three-dimensional forest vegetation structure. The objective of the study was to develop new methods for tree-level biomass estimation using metrics derived from ALS point clouds and to compare the results with field references collected using destructive sampling and with existing biomass models. The study area was located in Evo, southern Finland. ALS data was collected in 2009 with pulse density equalling approximately 10 pulses/m². Linear models were developed for the following tree biomass components: total, stem wood, living branch and total canopy biomass. ALS-derived geometric and statistical point metrics were used as explanatory variables when creating the models. The total and stem biomass root mean square error per cents equalled 26.3% and 28.4% for Scots pine (Pinus sylvestris L.), and 36.8% and 27.6% for Norway spruce (Picea abies (L.) H. Karst.), respectively. The results showed that higher estimation accuracy for all biomass components can be achieved with models created in this study compared to existing allometric biomass models when ALS-derived height and diameter were used as input parameters. Best results were achieved when adding field-measured diameter and height as inputs in the existing biomass models. The only exceptions to this were the canopy and living branch biomass estimations for spruce. The achieved results are encouraging for the use of ALS-derived metrics in biomass mapping and for further development of the models.     

Individual tree crown delineation using localized contour tree method and airborne LiDAR data in coniferous forests

Individual tree crown delineation is of great importance for forest inventory and management. The increasing availability of high-resolution airborne light detection and ranging (LiDAR) data makes it possible to delineate the crown structure of individual trees and deduce their geometric properties with high accuracy. In this study, we developed an automated segmentation method that is able to fully utilize high-resolution LiDAR data for detecting, extracting, and characterizing individual tree crowns with a multitude of geometric and topological properties. The proposed approach captures topological structure of forest and quantifies topological relationships of tree crowns by using a graph theory-based localized contour tree method, and finally segments individual tree crowns by analogy of recognizing hills from a topographic map. This approach consists of five key technical components: (1) derivation of canopy height model from airborne LiDAR data; (2) generation of contours based on the canopy height model; (3) extraction of hierarchical structures of tree crowns using the localized contour tree method; (4) delineation of individual tree crowns by segmenting hierarchical crown structure; and (5) calculation of geometric and topological properties of individual trees. We applied our new method to the Medicine Bow National Forest in the southwest of Laramie, Wyoming and the HJ Andrews Experimental Forest in the central portion of the Cascade Range of Oregon, U.S. The results reveal that the overall accuracy of individual tree crown delineation for the two study areas achieved 94.21% and 75.07%, respectively. Our method holds great potential for segmenting individual tree crowns under various forest conditions. Furthermore, the geometric and topological attributes derived from our method provide comprehensive and essential information for forest management.     

地基雷达的微波面散射模型对比与土壤水分反演

为了探究地基合成孔径雷达(cGBSAR)后向散射信号的时空变化规律和研究雷达土壤水分反演的影响因素,在内蒙古闪电河流域的昕元牧场站进行了地基雷达观测试验,本文结合以上观测试验的地基雷达数据进行波段、入射角度、极化通道3个雷达参数以及地表粗糙度参数对雷达的后向散射系数影响的分析,然后利用以上分析结果选择地表微波面散射模型...     

结合PCNN的红外与可见光图像多尺度融合

针对红外与可见光图像融合对比度不高,易丢失细节信息等问题,提出了一种非下采样Contourlet变换域内基于特征激励的自适应PCNN红外与可见光图像融合方法。PCNN模型采用平均梯度和赋时矩阵来自适应调节其链接强度和迭代次数等参数。对NSCT多尺度多方向高低频子带系数,分别采用特征激励PCNN,根据点火时间图的区域能量来选择融合系数。实验结果表明,该方法能够有效地融合红外和可见光图像信息,对比度高,细节保持好,无论在视觉效果上还是客观评价指标上,优于常用的图像融合方法。     

小波神经网络模型在高铁路基沉降预测中的应用研究

为了提高变形监测数据预测的精度与可靠性,提高神经网络预测方法的稳定性,尝试将小波分析与BP神经网络相结合的小波神经网络应用于高铁路基处的沉降监测数据处理中。综合小波分析与神经网络算法的优点,建立松散型及紧致型小波神经网络预测分析模型。通过实验数据对比分析,验证了采用紧致型小波神经网络预测模型能够较好地用来处理路基的动态变形监测数据,预测稳定性及预测精度较高。     

基于BP神经网络的霍普菲尔德模型改进研究

为了分析对流层延迟的时空变化规律、提高对流层延迟的改正精度,利用BP神经网络处理非线性问题的优势,改进传统的霍普菲尔德模型得到一种新的融合模型(Hop+BP模型)。分别对比Hop+BP模型与传统的霍普菲尔德模型、多元线性回归模型、BP神经网络等模型的计算结果,得到如下结论:霍普菲尔德模型存在一个明显的系统误差,精度较低;多元线性回归的预测精度有所提高,但是其本质是将数据强制拟合,缺少物理解释,难以推广使用;传统的BP神经网络的计算精度较之霍普菲尔德模型有80%的提高,但存在明显的不稳定性;Hop+BP模型具有预测精度高、稳定性好等优点,预测中误差为1.1cm,明显优于传统方法。     

车载LiDAR点云相连行道树精细分割

针对车载激光雷达获取的行道树树冠点云相连问题,该文提出了基于树木生长模型的距离加权分割方法。首先通过等距原则将相连树粗分割为单株树木,采用求差和拟合算法获得树木高度、胸径等属性信息;再根据树木的理论和实际生长模型,计算得到树冠距离加权值;最后在此基础上对树冠进行迭代处理,实现邻接行道树点云的精确分割。经过不同树种试验验证,该算法能够实现对多棵相连树点云的进一步精确分割。     

机器学习法在面向对象影像分类中的对比分析

针对如何选择遥感影像面向地理对象分类方法的问题,该文面向地理国情普查中的地表覆盖分类应用,以3个典型区域(山区、平原、城区)的多源高分辨率遥感影像为实验数据,从分类效果、分类精度等方面对比分析3种分类方法(支持向量机、决策树、随机森林)的优劣。在相同影像分割、特征提取、样本采集条件下,通过333组分类实验,得出以下规律:支持向量机分类方法稳定性强,分类速度快,但对特征数的要求高,特征数目与总体精度、地物环境之间的规律性不强,从而增加了特征提取与选择的难度,而随着特征的增加,决策树、随机森林的总体分类精度均为先升高后降低,最后趋于平衡。最后,综合随机森林对特征的优选机制和支持向量机的高分类精度,得到新的组合分类器。     

A spatial–temporal Hopfield neural network approach for super-resolution land cover mapping with multi-temporal different resolution remotely sensed images

The mixed pixel problem affects the extraction of land cover information from remotely sensed images. Super-resolution mapping (SRM) can produce land cover maps with a finer spatial resolution than the remotely sensed images, and reduce the mixed pixel problem to some extent. Traditional SRMs solely adopt a single coarse-resolution image as input. Uncertainty always exists in resultant fine-resolution land cover maps, due to the lack of information about detailed land cover spatial patterns. The development of remote sensing technology has enabled the storage of a great amount of fine spatial resolution remotely sensed images. These data can provide fine-resolution land cover spatial information and are promising in reducing the SRM uncertainty. This paper presents a spatial–temporal Hopfield neural network (STHNN) based SRM, by employing both a current coarse-resolution image and a previous fine-resolution land cover map as input. STHNN considers the spatial information, as well as the temporal information of sub-pixel pairs by distinguishing the unchanged, decreased and increased land cover fractions in each coarse-resolution pixel, and uses different rules in labeling these sub-pixels. The proposed STHNN method was tested using synthetic images with different class fraction errors and real Landsat images, by comparing with pixel-based classification method and several popular SRM methods including pixel-swapping algorithm, Hopfield neural network based method and sub-pixel land cover change mapping method. Results show that STHNN outperforms pixel-based classification method, pixel-swapping algorithm and Hopfield neural network based model in most cases. The weight parameters of different STHNN spatial constraints, temporal constraints and fraction constraint have important functions in the STHNN performance. The heterogeneity degree of the previous map and the fraction images errors affect the STHNN accuracy, and can be served as guidances of selecting the optimal STHNN weight parameters.