喀斯特山区居民地多尺度遥感信息提取精度对比

发布时间：2018-04-02 09:27

本文选题：喀斯特　切入点：居民地　出处：《中国岩溶》2017年04期

【摘要】：以贵州省典型喀斯特山区晴隆县为研究区,以分辨率分别为2.1m的资源三号(ZY3)影像、10m的哨兵二号(Sentinel-2)影像、16m的高分一号(GF1)影像以及30m的Landsat8影像为数据源,利用面向对象分类和人机交互相结合的方法提取研究区城市居民地、乡镇居民地和农村居民地,并将4种不同分辨率影像上提取的3种居民地信息与从0.5m分辨率的普莱亚(Pleiades)影像通过目视解译且经实地核查得到的数据进行对比,计算出4种分辨率影像上3种居民地信息的提取精度,对提取结果进行精度对比分析。结果表明:(1)同一分辨率影像中,城镇居民地提取误差小于农村居民地;不同分辨率影像中,随影像分辨率大小的降低3种居民地中城市居民地的精度变化最小,为23.99%,农村居民地的数据精度变化最大,达到35.3%。(2)从2.1m到30m分辨率影像,居民地信息错分总误差快速增加,总误差比依次为:2.56%、15.58%、24.50%、32.72%,城市居民地错分误差比明显小于农村居民地,且3种居民地错分为其他地类面积最多的均是裸地;(3)随着影像分辨率大小的降低,居民地漏分总误差分别为2.86%,18.60%,27.99%,37.49%,其中分散式的农村居民地识别更易受到周围环境的影响,漏分误差随影像分辨率大小降低而显著增加,4种分辨率影像中3种居民地信息漏分误差最小的均为水体,最大的是裸地和耕地。
[Abstract]:Taking Qinglong County, a typical karst mountain area of Guizhou Province, as the research area, the data source is the Sentinel-2 Sentinel-2 image with a resolution of 2.1 m and the Sentinel-2 image with a resolution of 2.1 m, and the Landsat8 image with a resolution of 30 m. The method of combining object oriented classification and human-computer interaction is used to extract urban, township and rural residential land in the research area. At the same time, the three kinds of residential information extracted from 4 different resolution images were compared with those obtained from 0.5 m resolution Pleiades images, which were visually interpreted and verified in the field. The extraction accuracy of three kinds of resident land information on 4 kinds of resolution images was calculated and compared and analyzed. The result shows that in the same resolution image, the extraction error of urban land is smaller than that of rural land. Among the different resolution images, with the decrease of the resolution of the image, the precision of the urban land is the least, which is 23.99, and the data precision of the rural land has the greatest change, reaching 35.33.32) from 2.1 m to 30 m. The total error of misdivision of residential land information increased rapidly, and the total error ratio was in turn:: 2.56 / 15.58 / 24.50 / 32.72. The error ratio of urban residents was obviously smaller than that of rural areas. With the decrease of the resolution of the image, the total error of the local floor drain is 2.866.600.27.99. among them, the decentralized identification of the rural residents is more susceptible to the influence of the surrounding environment. The leakage error increased significantly with the decrease of the resolution of the image. Among the three kinds of resolution images, the smallest error was water, the largest was bare land and cultivated land.
【作者单位】：贵州师范大学地理与环境科学学院/贵州省山地资源与环境遥感应用重点实验室;
【基金】：国家自然科学基金(41361091) 贵阳市白云区科技计划项目(白科合同[2016]61号)
【分类号】：P237

【相似文献】