摘要: |
[目的]为探讨多源中等分辨率数据在冬小麦种植时空分布上的应用。[方法]利用2017年冬小麦关键生育期的Landsat8 OLI(抽穗期)和时间序列的GF 1 WFV(2016—2017生育期)数据,在分析各个行政分区的地表覆盖状况、作物结构和地块破碎度差别的基础上,将行政区划分为3种类型不同的提取单元并建立了适合于各自分区的提取模型:(1)利用关键生育期的OLI数据,采用监督分类—神经网络方法提取结构单一、地块齐整的怀远县种植区; (2)基于WFV数据构建五河县及城区种植区的冬小麦全生育期NDVI时间序列曲线,根据NDVI的时间特征构建冬小麦提取的决策树分类模型提取结构较复杂、混合像元明显的五河县及城区种植区; (3)在对关键生育期OLI NDVI数据合理分割的基础上,采用最大似然的面向对象分类法获取种植密集、地块破碎的固镇县种植区。[结果]提取结果采用混淆矩阵和当年度统计数据相结合的方法进行精度评价,结果表明:(1)怀远县提取出的冬小麦提取总体精度为9791%,五河县及城区提取出的精度为9762%,固镇县的精度为9742%; (2)全区域冬小麦提取的总体精度为8682%,Kappa系数为084。与当年度统计数据对比的结果表明: 2017年蚌埠市的准确提取面积精度可达9791%,提取面积数据小于蚌埠市统计年鉴提供的统计数据,与调查的实际种植地块基本一致。[结论]采用不同方法提取不同空间分布特征的冬小麦种植面积具有较好的精度,该方法可以为市域冬小麦面积提取提供技术参考。 |
关键词: 冬小麦多时相神经网络决策树面向对象方法 |
DOI: |
分类号:S127 |
基金项目:安徽高校人文社会科学研究重点项目(SK2017A0574); 安徽科技学院引进人才资助项目(ZHYJ201601,ZHYJ201603) |
|
EXTRACTION OF WINTER WHEAT PLANTED AREAS FROM LANDSAT 8 AND GF 1 DATA SATELLITE IMAGERY USING A HYBRID METHOD |
Kan Zhiyi, Hu Huijuan, Liu Jikai, Li Xinwei, Zhang Wei
|
University of Science and Technology of Anhui, Fengyang, Anhui 233100, China
|
Abstract: |
Comprehensive utilization of multi source and time series remote sensing data in crop classification and extraction has become a hot topics in recent years. Compared with other data sources, the medium resolution data have been widely used because of its high precision, time sensitive and easy access. We divided the study area into three parts, i.e. Huaiyuan county, Wuhe county and Guzhen county under the constraint of administrative boundary. Huaiyuan county has a neat land of a single structure. Wuhe County and its urban areas have complex land structures, thereby reflecting mixed pixels in the Satellite imagery. Guzhen county includes the dominant patchy planting areas from which crops are densely planted. To investigate the spatial and temporal distributions of winter wheat, by considering planting structure of crops, topography and land fragmentation, we established the different mathematical models according to the above three countries. Factoring in key growth period of winter wheat in 2017 Landsat 8 OLI (heading stage) and the time series of GF 1 WFV (growth period in 2016-2017) and the information about planting structure, geomorphology and fragmentation of landmass. Using critical growth periods of OLI data , the artificial neural network had been used to extract the distribution of winter wheat in Huaiyuan County. Using NDVI time series curve of the whole growth period of winter wheat in Wuhe county and its urban areas, the decision tree model was built to extract the distribution of winter wheat in Wuhe County. For Guzhen country, we used object oriented and Maximum likelihood method that required segmented patches of OLI NDVI imagery as inputs to obtain distributions of winter wheat. The results showed that segmented precision of winter wheat for Huaiyuan, Wuhe and Guzhen county reached up to 97.91%, 97.62% and 97.42%, respectively . The overall accuracy and Kappa coefficient for evaluation of winter wheat in these three countries were 86.82%, and 0.84, respectively. Compared with statistical data in the current year showed that the accuracy of the accurate extraction area in Bengbu city in 2017 reached 97.91%, which was slightly smaller than statistical datum provided by Bengbu Statistical Yearbook. We can make safe conclusion ensures that the proposed method fully considers the differences in terms of planting structure of crops, topography and land fragmentation, and ensure the extraction accuracy of the winter wheat. |
Key words: winter wheat multi season Artificial Neural Network decision tree object based method |