顾及PolSAR影像极化-空间信息的农作物识别方法

A novel crop classification method based on polarimetric and spatial characteristics of PolSAR image

针对传统基于全极化合成孔径雷达(PolSAR)影像的农作物分类方法因不同农作物的极化、纹理信息描述不准确及特征冗余,导致农作物识别方法低精度与低效率的问题,该文在基于优选的极化特征基础上,通过综合利用不同农作物在PolSAR影像表现的空间上下文信息,形成对应纹理特征,并融合随机森林分类器与马尔可夫随机场算法,提出一种顾及PolSAR影像极化-空间信息的农作物类型识别新方法。基于荷兰Flevoland省及海南陵水县定性与定量结果表明:与传统农作物分类方法比较,本文有效解决农作物类型识别过程受极化-纹理特征构建不准确问题,显著提高了农作物识别的总体精度与Kappa系数(两实验区总体精度分别提升了5.5%和2.6%,Kappa提升了0.061和0.037),为农业相关部门决策提供技术支持。

The traditional crop classification methods have not used well the polarimetric and spatial features of PolSAR(fully polarimetric synthetic aperture radar)images and the classification accuracy is not ideal due to the continuous improvement of spatial resolution of PolSAR images.In view of the shortcomings of traditional crop classification methods,a novel crop classification method based on polarimetric and spatial characteristics of PolSAR image was proposed in this paper,which was constructed texture features,Random Forest(RF)classifier,and Markov random field(MRF)to realize the accurate recognition of crop types,which can make full use of the polarimetric and spatial information of PolSAR images.The experiments are implemented with data sets of Flevoland and Lingshui.In comparison to state-of-the-art crop classification methods,the proposed crop classification method performs better in both qualitative and quantitative evaluations.Meanwhile,the overall accuracy and Kappa of crop classification were significantly improved when compared with other crop classification methods and the feasibility and robustness of the proposed method were verified(The OA of crop classification are significantly improved 5.5%and 2.6%,meanwhile,Kappa coefficient are improved 0.061 and 0.037),which can provide the technical support for decision-making of relevant agricultural departments.