• 主题
高级搜索  >
历史搜索 清空历史
首页> 外文会议>Asian conference on remote sensingACRS >APPLICATION OF OBJECT-BASED IMAGE ANALYSIS AND SUPPORT VECTOR MACHINE TO CREATE BROADSCALE MAP OF SEAGRASS USING LANDSAT 8 IMAGE: A CASE STUDY IN CALATAGAN, BATANGAS
【24h】

APPLICATION OF OBJECT-BASED IMAGE ANALYSIS AND SUPPORT VECTOR MACHINE TO CREATE BROADSCALE MAP OF SEAGRASS USING LANDSAT 8 IMAGE: A CASE STUDY IN CALATAGAN, BATANGAS

机译：基于对象的图像分析和支持向量机的应用使用Landsat 8拍摄海草的广场地图8图片：熊塘斯卡拉塔根的案例研究

获取原文
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

As an invaluable part of the benthic habitat, the assessment of seagrass meadows is required in coastal monitoring and conservation. This paper presents a method in the identification of seagrass using Landsat 8 image applied in the shallow water areas of Calatagan, Batangas. Radiometric calibration, atmospheric correction (Fast Line-of-sight Atmospheric Analysis of Hypercubes) and pansharpening (Gram-Schmidt) were applied in the raw Landsat 8 OLI/TIRS (Operational Land Imager/Thermal Infrared Sensor) image. The traditional approach of pixel-based classification in Landsat image was replaced by Object-based Image Analysis performed in eCognirion. Using multiresolution segmentation with scale = 0.3, shape = 0.1 and compactness = 0.1, the best segmentation output was produced giving weights to coastal aerosol, blue, green, red and near infrared bands. The study investigated the application of Support Vector Machine (SVM) to discriminate seagrass in a broadscale manner. For the training phase of the SVM, 12 vegetation indices (DVI, EVI, GDVI, GNDVI, GRVI, NDVI, NG, OSAVI, RVI, SAVI, VARIgreen and modified NDVI) and three spectral bands (NIR, SWIR1 and SWIR2) were used. Performing statistical analysis on the vegetation indices for the dense seagrass training samples, NDVI and SAVI showed higher values of coefficient of variation (＞ 38%), while NG, OSAVI and modified NDVI obtained lower values of coefficient of variation (＜ 9%). Removing NDVI and SAVI as input features had very minimal effects on improving the accuracy. We classified images on the shallow water according to dense seagrass, sparse seagrass and non-seagrass (which includes mangrove, cloud, sea water and sand). The results indicate that using SVM on the segmented objects can accurately map dense seagrass cover [PA = 0.92, UA = 0.98]. Visual inspection of the produced dense seagrass cover map (15-m resolution) on the available 0.5-m resolution orthophotograph produced agreeable results. The confusion among sparse seagrass [PA = 0.85, UA = 0.97], dense seagrass, and non-seagrass was apparent because some sparse seagrass cover are dominated by sand; and, it is quite hard to quantify the percentage of the seagrass cover in every pixel of the Landsat 8 image even if it was pan-sharpened. The overall accuracy of the three-scheme classification records 0.97 with Kappa Index of Agreement of 0.92. Further improvements on classifying seagrass cover submerged in relatively deeper water can still be studied together with advanced techniques such as data mining to determine the optimal value of the cost parameter (C) and kernel spread function (γ) as SVM parameters for improved classification accuracy.
机译：作为终身栖息地的宝贵部分，沿海监测和保护需要对海草草甸的评估。本文介绍了使用Landsat 8图像识别海草的方法，应用于巴拉瓜，巴拉邦加的浅水区。辐射校准，大气校正（超速度快速的瞄准型大气分析）和泛散纹（克施密特）应用于原始Landsat 8 Oli / Tirs（运营陆地成像器/热红外传感器）图像。通过在认知中进行的基于对象的图像分析取代了旧岩地图像中基于像素的分类的传统方法。使用刻度= 0.3的多分辨率分割，形状= 0.1和紧凑率= 0.1，产生最佳分割输出，为沿海气溶胶，蓝色，绿色，红色和近红外线带提供重量。该研究调查了支持向量机（SVM）以广泛的方式辨别海草。对于SVM的训练阶段，12个植被指数（DVI，EVI，GDVI，GNDVI，GRVI，NDVI，NG，OSAVI，RVI，SAVI，VARIgreen和改性的NDVI）和三个光谱带（NIR，SWIR1和SWIR2）被用来。对密集的海草训练样品的植被指数进行统计分析，NDVI和SAVI显示出更高的变异系数（> 38％），而NG，奥萨维生和改性NDVI获得的变异系数的较低值（<9％）。删除NDVI和Savi作为输入功能对提高准确性具有很小的影响。根据密集的海草，稀疏的海草和非海草（包括红树林，云，海水和沙子），我们在浅水上进行分类图像。结果表明，在分段对象上使用SVM可以精确地映射致密的海草盖[PA = 0.92，UA = 0.98]。在可用的0.5米分辨率正方摄光镜的可用致密海草覆盖图（15米分辨率）的目视检测产生令人愉快的结果。稀疏海草的混乱[Pa = 0.85，UA = 0.97]，密集的海草和非海草是显而易见的，因为一些稀疏的海草覆盖物由沙子主导;并且，即使泛尖锐化，也很难量化Landsat 8图像的每个像素中的海草盖的百分比。三个方案分类记录的总体准确性为0.97，kappa协议达0.92。上分类海草盖在相对较深水淹没的进一步改进仍然可以连同先进的技术，如数据挖掘，以确定成本参数（C）和内核扩展函数（γ）作为SVM参数以提高分级精度的最优值的影响。

著录项

相似文献

  • 外文文献
  • 专利
获取原文

客服邮箱：kefu@zhangqiaokeyan.com

京公网安备：11010802029741号 ICP备案号：京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有

  • 客服微信

  • 服务号