黄冰晶+谢明鸿+李润青
摘要:传统的遥感影像目标检测方法大多利用人工提取特征,难以用于背景复杂的高分辨率遥感影像。以高分辨率遥感影像建筑物图像为研究对象,设计了一种基于LVQ神经网络的建筑物提取方法。对图像提取其颜色、纹理与形状特征,构成图像特征矢量并将其特征归一化,利用LVQ神经网络识别并提取出建筑物。通过与其它典型神经网络识别方法进行实验比较,结果表明该算法相对于单层感知器识别率提高了10.0%,比BP神经网络识别率提高了22.5%,能取得更理想的提取效果。
关键词:高分辨率遥感影像;建筑物提取;LVQ神经网络;特征提取
DOIDOI:10.11907/rjdk.171955
中图分类号:TP317.4
文献标识码:A 文章编号:1672-7800(2017)012-0197-05
Abstract:Traditional detection methods of remote sensing images mostly use the feature of artificial extraction and are difficult to be applied to high-resolution remote sensing images with complex backgrounds. Based on the high resolution remote sensing of the image building, we proposed a building extraction method based on LVQ neural network. First,extract the feature of the color, texture and shape, as the three combine to form the feature vector of the image and be normalized; second, utilize the LVQ neural network to identify and extract the buildings. Compared with other typical neural network identification methods, the results show that the proposed algorithm can achieve more ideal extraction effect as the recognition rate of LVQ neural network is 10.1 percentage points higher than that of the single layer sensor and 22.5 percentage points higher than that of the BP neural network.
Key Words:high-resolution remote sensing image; building extraction; LVQ neural network; feature extraction
0 引言
近年,遙感技术迅猛发展推动了遥感影像空间分辨率快速提高,高分辨率影像在遥感技术应用数据来源中占很大比重。建筑物作为遥感影像一类重要且具有显著特征的目标,在测绘、城市规划、军事侦察、打击效果评估等军事与民用领域具有重要作用,研究高分辨率遥感影像建筑物自动提取技术具有现实意义与理论价值[1-3]。城区建筑物提取是目前研究重点,如何从高分辨率遥感影像中进行高精度建筑物提取成为主要研究内容。
基于遥感影像的建筑物提取,20多年来学者做了大量研究。Dragut L[4]等使用eCognition软件研究了多尺度分割的自动参数化,并应用于建筑物分割与提取,取得了较高提取精度。该方法优点是能有效减少椒盐效应影响,减小了类内方差,能有效利用影像结构、形状属性,可与GIS系统结合得到矢量化输出结果;……