OIF辅助BP神经网络模式识别的土地覆盖分类  

Land cover classification based on OIF assisted BP neural network pattern recognition

作　　者：刘瑞航 张佳瑜 马彩虹 LIU Ruihang;ZHANG Jiayu;MA Caihong(School of Geography and Planning,Ningxia University,Yinchuan Ningxia 750021,China)

机构地区：[1]宁夏大学地理科学与规划学院,宁夏银川750021

出　　处：《北京测绘》2022年第9期1145-1150,共6页Beijing Surveying and Mapping

基　　金：国家自然科学基金项目(41961034);宁夏自然科学基金项目(2021AAC03087,2020AAC03053);宁夏高等学校科研项目(NGY2020012)。

摘　　要：精准高效的土地覆盖分类不仅能提高土地利用类型研究与制图质量,也有助于有序地组织土地的保护与开发。以2020年7月28日银川市三区的美国陆地卫星8号陆地成像仪(Landsat8 OLI)影像为数据源,建立最佳指数法(OIF)辅助前馈(BP)神经网络的模式识别模型,借助遥感影像处理平台(ENVI)选取训练样本集,对银川市三区土地覆盖类型进行分类。结果表明,OIF辅助BP神经网络模式识别的总体精度为86.95%,Kappa系数为0.8435,优于最大似然法和支持向量机(SVM)分类法;从生产者精度来看,该方法在林地、草地、水域和未利用地的分类中优势明显,精度都超过了90.00%,但在建设用地的分类中精度较低,只有70.73%。实验证明,OIF辅助BP神经网络的模式识别在遥感影像的监督分类中是一种精准高效的分类方法。Accurate and efficient land cover classification can not only improve the quality of land use type research and mapping,but also help to orderly organize land protection and development.Taking the landsat8 operational land imager(OLI)image of three districts of Yinchuan City on July 28,2020 as the data source,the pattern recognition model of optimum index factor(OIF)assisted BP neural network was established,and the training sample set was selected with the help of ENVI remote sensing image processing software to classify the land cover types of three districts of Yinchuan city.The results showed that the overall accuracy of OIF assisted BP neural network pattern recognition was 86.95%and Kappa coefficient was 0.8435,which was better than Maximum Likelihood method and support vector machine(SVM)classification method;From the perspective of producer accuracy,this method had obvious advantages in the classification of forest land,grassland,water area and unused land,with an accuracy of more than 90.00%,but the accuracy in the classification of construction land was low,only 70.73%.Experiments showed that OIF assisted BP neural network pattern recognition was an accurate and efficient classification method in the supervised classification of remote sensing images.

关 键 词：最佳指数法 前馈神经网络 模式识别 遥感影像分类 土地覆盖 

分 类 号：P258[天文地球—测绘科学与技术] TP79[自动化与计算机技术—检测技术与自动化装置]