基于SAR遥感的北方旱地秋收作物识别研究

发布时间：2019-03-19 09:17

【摘要】：在我国北方旱地秋收作物生长关键期,云雨天气影响较大,无法及时、有效地获取光学遥感数据,因此利用雷达遥感进行旱地作物识别研究非常必要。文章以河北省衡水市为研究区,选择6期RADARAST-2全极化影像作为数据源,分类方法为随机森林法。首先通过对比不同时相间的组合结果,优选出了研究区典型秋收作物(玉米、棉花)的最佳识别时相及组合方式。其次,提取最优识别时相的后向散射信息、纹理信息、极化分解等3部分信息,依据信息间相互组合的结果及随机森林算法对变量的重要性评价,文中对上述3部分信息进行了重要性评估。结果表明:利用SAR识别旱地秋收作物时应着重关注作物生长前期的时相,其中玉米在6月27日单一时相下就可获得90%以上的高精度;棉花面积小、地块破碎,但通过6月3日与6月27日两个时相的结合也得到了70%以上的精度。在玉米识别中极化信息具有较大的贡献,极化变量的加入主要增加了玉米和建筑用地的可分离性,与单纯利用后向散射信息分类相比精度提高了近7%;同样,纹理信息和极化分解信息的加入也使棉花的精度提高了3%。最后,利用随机森林算法对变量的重要性评价机制,优选出对玉米识别最为重要的5个变量,依次为:VH、Alpha、Yamaguchi4-Odd、Freeman-Vol和Mean(HV)。该研究利用雷达数据进行旱地作物识别,验证了雷达影像对旱地秋收作物的识别能力,不仅保证了数据获取与天气状况的独立性,还凭借SAR独有的数据获取方式,为光学数据提供了补充。
[Abstract]:In the key period of autumn crop growth in the north of China, cloud and rain weather have great influence on crop growth, so it is very necessary to use radar remote sensing to identify crops in drylands because it is difficult to obtain optical remote sensing data in time and effectively. In this paper, Hengshui City, Hebei Province is chosen as the study area, 6-phase RADARAST-2 polarimetric images are selected as the data source, and the classification method is the random forest method. Firstly, by comparing the results of different interphase combinations, the optimal identification time phase and combination mode of typical autumn harvest crops (maize, cotton) in the study area were optimized. Secondly, we extract the backscattering information, texture information and polarization decomposition information of the optimal identification phase, and evaluate the importance of random forest algorithm to the variables according to the results of the combination of the information and the random forest algorithm. In this paper, the importance of the above three parts of information is evaluated. The results showed that when using SAR to identify the crops in dry land, we should pay more attention to the early phase of crop growth, in which maize could get more than 90% high precision under the single phase on June 27. The cotton area is small and the block is broken, but through the combination of June 3 and June 27, more than 70% precision has been obtained. Polarization information plays an important role in maize recognition. The polarization variable mainly increases the separability of maize and construction land, and the precision is improved by 7% compared with the classification of backscatter information. Similarly, the addition of texture information and polarization decomposition information also increased the accuracy of cotton by 3%. Finally, using the stochastic forest algorithm to evaluate the importance of variables, the five most important variables for maize identification are selected, which are: VH,Alpha,Yamaguchi4-Odd,Freeman-Vol and Mean (HV). This study uses radar data to identify dryland crops, validates the ability of radar images to identify dryland autumn crops, not only ensures the independence of data acquisition and weather conditions, but also relies on the unique data acquisition method of SAR. It provides a supplement to the optical data.
【作者单位】：中国农业科学院农业资源与农业区划研究所;农业部农业信息技术重点实验室;
【基金】：国家科技重大专项项目“高分农业遥感监测与评价示范系统”(09-Y30B03-9001-13/15)
【分类号】：S127

【参考文献】