基于地块尺度多时相遥感影像的冬小麦种植面积提取

针对仅利用单一遥感影像数据获取农作物信息精度不够问题，该文选择冬小麦主产地河南省兰考县乡镇作为研究区，以2017年多时相中分辨率Landsat8 OLI影像和Google earth上下载的亚米级高分影像为遥感数据源，结合光谱差异和农田地块信息实现冬小麦的精确提取。该算法首先构建不同时相决策树模型，分别实现2个时相的冬小麦区域初步提取；其次通过将对高分影像多尺度分割产生的地块信息分别与2个时相冬小麦播种面积初步区域相互叠加，完成地块单元控制下的冬小麦播种面积分地块统计，并通过设定不同统计阈值，分析落在每一地块单元下的冬小麦区域，生成基于地块单元的冬小麦播种面积分布图；最后通过多时相交叉验证，获取最终冬小麦播种区域。结果表明：该方法能更加准确提取冬小麦种植面积，保持较低的误判率（1.3%）水平下，得到较高的提取正确率（95.9%），较通过对比单一Google earth高分辨率影像获取冬小麦精度（85.6%）高，该研究对通过融合多源多时相影像数据获取农作物提供参考。

Winter wheat planting area extraction using multi-temporal remote sensing images based on field parcel

Abstract: The estimation of winter wheat area based on remote sensing images is one of important contents in filed of agriculture information monitoring. However, it is difficult to solve the problems of spectrum heterogeneity in the same farmland and spectra similarity between different farmlands timely and accurately using mid-resolution images. In order to maximally avoid problems metioned above and accurately map the planting area of winter wheat, a object-parcel classification method was developed in the study area of Lankao Cunty, Henan Province. An improved identification procedure for geo-parcel based winter wheat identification was presented, combining fine-resolution image and multi-temporal medium-resolution images. Combined spectrum and filed parcel information, precisely extraction of winter wheat planting area was realized from multi-temporal OLI images and Google earth high-resolution images (resolution of 0.49 m) through the following several steps. 1) Constructing winter wheat decision tree extraction models to extract the simplified winter wheat area based on spectral feature. Crops performed different phenological characteristics during the growth and development stage, which displayed spectral differences on remote sensing images. And to obtain the optimum temporal phases to extract winter wheat planting area, temporal phase among typical crops in study area was analyzed based on the phenological characteristics; 2) The field parcel information generated from high-resolution imagery by multi-scale segmentation algorithm. And then, according to the field parcels obtained on the high-resolution images, the two simplified OLI images of winter wheat were superimposed on the parcel respectively. Partition statistics ratio (proportion of simplified winter wheat in each field parcels) was calculated, and then the winter wheat parcels on the high-resolution images were obtained based on partition statistics ratios. Finally, analyzing the extraction accuracy under different statistics ratio threshold, then generating high-resolution winter wheat plots based on the parcel; 3) Through cross validation, the winter wheat planting area was extracted. Identification results of the winter wheat with the parcel statistics ratio threshold of 0.20 in the phase-1 (OLI image on 2017-03-04, with higher extraction correctness ratio and lower misjudgment ratio) and the recognition result with the phase-2(OLI image on 2017-05-07) threshold of 0.30 were selected for cross-validation. The experiment result showed that the method could recognize winter wheat area accurately. The higher recognition accuracy (95.9%) was obtained under the lower misjudgment ratio (1.3%). Last but not least, an application of proposal method in Lankao County was performed to verify the accuracy of winter wheat extraction with the correctness ratio of 91.5%. And the accuracy of winter wheat recognition could be expected higher in regions with simple planting structure or less fragmental parcels. The omission of winter wheat extraction based on per-parcel classification mostly happened in the fragmental parcels, coupled with the accuracy of segmentation, because the parcels were not completely segmented according to the single crop type. Finally the performance of partition statistics ratio analysis in distinguishing pure winter wheat parcels and mixed winter wheat parcels was tested by controlling the partition statistics threshold. The identification results indicated that the integration of high spatial-temporal resolution imagery is promising for crop identification based on geo-parcel .