基于最优尺度的遥感影像土地覆盖分类仿真
Land Cover Classification Method of Remote Sensing Image Based on Optimal Scale
李晨睿 1赖雨诗 1吴燕杰 1夏召强2
作者信息
- 1. 首都师范大学资源环境与旅游学院,北京 100048
- 2. 西北工业大学电子信息学院,陕西 西安 710029
- 折叠
摘要
土地覆盖遥感影像是国家的战略性、基础性资源,真实、准确和实时的土地覆盖类型信息对科学保护和合理利用土地资源至关重要.随着大数据时代遥感影像数量快速增长,已有算法的准确性和稳定性无法满足土地覆盖情况分类需求.为进一步提升土地覆盖分类准确率,提出一种基于最优尺度分割与特征融合的方法.首先针对预处理后的遥感影像,利用局部方差计算出分割的最优尺度,并以尺度为基准优化过分割、欠分割影像;然后以分割后的影像为基准,采用局部二值模式算子(LBP)及神经网络提取土地影像的纹理特征和光谱初级特征;最后将影像的两种特征有机融合,并利用支持向量机分类器(SVM),构建了土地遥感影像分类模型(OSF-SVM模型).分割仿真结果表明,与已有方法相比,文中的尺度分割技术在RR、RI及ARI指标上具有所提高,平局提升了10.83%;分类仿真结果表明,较传统SVM模型相比,OSF-SVM模型在R、P以及F1 指标上分别平均提高了4.1%、3.9%和4%.因此,通过最优尺度分割和特征融合构建的OSF-SVM遥感影像土地覆盖分类模型,提高了影像分割及分类的精确度与稳定性.
Abstract
Remote sensing images of land cover is a strategic basic resource of the country.Real,accurate and re-al-time information of land-cover type is crucial to scientific protection and rational utilization of land resources.In the era of big data,the number of remote-sensing images is growing rapidly,and the accuracy and stability of existing algorithms cannot meet the needs of land-cover classification.This paper proposes a method based on the combination of optimal scale segmentation and feature fusion.The method comprises the following steps.Firstly,an optimal scale for segmentation is calculated according to the local variance of a remote sensing image,and the over-segmented and under-segmented images are optimized by taking the scale as a reference.Secondly,the texture feature and the prima-ry feature of a land image are extracted by taking the segmented image as a reference with the local binary pattern(LBP)operator and neural network.Finally,after image preprocessing,the two features of the image are fused to be the input of a support vector machine(SVM)classifier as a base kernel.The classification model of land remote sensing image based on optimal scale segmentation and feature fusion(OSF-SVM model)is constructed.The simula-tion results of segmentation experiments show that,compared with the existing methods,the scale segmentation tech-nology in this paper has improved RR,RI and ARI indicators,and the average score has increased by 10.83%.The simulation results of classification experiments show that,compared with the traditional SVM model,the OSF-SVM model has an average increase of 0.041,0.039 and 0.040 in R,P and F1 indicators,respectively.Therefore,the re-mote sensing image land-cover classification model constructed in this paper improves the accuracy and stability of image segmentation and classification through optimal scale segmentation and feature fusion.
关键词
遥感影像/最优尺度/影像分类Key words
Remote sensing image/Optimal scale/Image classification引用本文复制引用
基金项目
北京市教委科学研究计划(KZ202210028045)
出版年
2024
NETL
NSTL
万方数据