资源简介

摘要： 在城镇景观中，场景噪声、阴影遮挡、植物群之间光谱相似等形成的负面影响，使传统的监督分类方法无法满足精度要求。为此，提出一种适用于城镇植物群的软分类方法。在常规BP网络监督分类的基础上，做了3处改进：①在特征空间堆叠冬、夏季图像特征，以增加特征空间维度，适应复杂分类。②以BP网络软输出集群原型，并据原型中确定成员的信息、递归推测模糊成员的类码，实现软分类。③根据软分类对象尺寸、类别、位置和彼此的邻近关系，滤除和吸收噪声图斑；以及将树冠本影补充到树冠对象中，而使植物群对象更完整、准确。MATLAB测试结果显示，在一个由32个描述符（2个季节的描述符堆叠，每个季节16个）组成的特征空间中，使用高空间分辨率卫星真彩色图像，可以对4种主要植物群类别和水体、其他背景分类。与硬分类、单季节的传统方法相比，新方法分类的全局精度（OA）和卡帕系数（κ）平均分别提高30.25%和40.61%。说明该法在城镇植物群遥感自动分类方面具有鲁棒和普适性。 关键词： 城镇植物群;  双季节;  BP网络;  解模糊;  对象邻近分析     Abstract: There are some negative effects on the classification of urban vegetation population from image noise，building shading and poor spectral separability between vegetation populations as in urban landscape.A soft classification method suitable for urban vegetation populations，referred to here as SCV(Soft Classification of Vegetation) has been proposed.Compared with conventional methods of supervised classification，there are some newly explored algorithms in SCV，such as adding double seasonal information to the classification feature space，deriving soft partition prototype by BP network and defuzzifying the prototype afterward and removing image noisy and supplying tree crown patches by adjacency analysis between objects.The classification is conducted in a feature space consisting of thirty\|two descriptors(sixteen for each season).Four categories of urban vegetation populations，the evergreen vegetation，deciduous vegetation，hydrophytes and grass land，can be determinately extracted from QuickBird true color images.Results show that(1) the adding of season information can improve separability between vegetation populations therefore increasing the overall accuracy(OA) of classification;(2) the soft partition prototype can be divided into sure and fuzzy member sets and then the latter can be relabeled through recursive defuzzifying;(3) the adjacency analysis between objects will make the extracted vegetation objects more complete.The tests by using the software of MATLAB indicate that this approach has better robustness and universality in the classification of urban vegetation populations.Average OA and Kappa coefficient(κ) are 87.4% and 83.1% respectively as using SCV.In contrast，average OA and κ are 67.1% and 59.1% respectively as taking a conventional classification by hard BP network and using only single seasonal data.However，the problem of separating different vegetation populations in shadowed scene still needs to be solved in future.

相关资源

1. 水质遥感监测的关键要素叶绿素a的反演算法研究进展
2. 基于无人机遥感技术的森林参数获取研究进展
3. 基于无线传感器网络的森林生态系统观测试验平台构建
4. 基于无人机激光雷达的森林冠层高度分析
5. 多源时序国产卫星影像的森林火灾动态监测
6. 基于无人机遥感技术的台风灾害倒伏绿化树木检测

下载说明

·如果您发现该资源不能下载，请通知管理员.gissky@gmail.com

·为确保下载的资源能正常使用，请使用[WinRAR v3.8]或以上版本解压本站资源,缺省解压密码www.gissky.net ，如果是压缩文件为分卷多文件，请依次下载每一个文件，并按照顺序命名为1.rar,2.rar,3.rar...，然后鼠标右击1.rar解压.

·为了保证您快速的下载速度，我们推荐您使用[网际快车]等专业工具下载.

·站内提供的资源纯属学习交流之用,如侵犯您的版权请与我们联系.