|
|||||
|
|
| GEE平台下利用物候特征进行面向对象的水稻种植分布提取 | |
| 引用本文： | 刘通, 任鸿瑞. GEE平台下利用物候特征进行面向对象的水稻种植分布提取[J]. 农业工程学报, 2022, 38(12): 189-196. DOI: 10.11975/j.issn.1002-6819.2022.12.022 |
| 作者姓名： | 刘通 任鸿瑞 |
| 作者单位： | 1.太原理工大学测绘科学与技术系，太原 030024 |
| 基金项目： | 中国气象局沈阳大气环境研究所联合开放基金（2021SYIAEKFMS39）；山西省重点研发计划(国际科技合作)项目（201903D421089） |
| 摘 要： | 为高效提取高精度水稻种植分布及其面积，该研究基于谷歌地球引擎（Google Earth Engine，GEE）平台，以辽宁省盘锦市为研究区，利用2020年Sentinel-2影像提取水稻生命周期内4个水稻物候期相应的光谱指数，利用简单非迭代聚类（Simple Non-Iterative Clustering，SNIC）算法来分割影像，灰度共生矩阵（Gray Level Co-occurrence Matrix，GLCM）来计算纹理特征，结合支持向量机（Support Vector Machine，SVM）和随机森林（Random Forest，RF）算法构建6种不同的模型进行水稻种植分布提取，并基于目视解译及实地调查数据，对比6种模型提取水稻的验证精度和实测精度，确定最优模型。结果表明：在水稻种植分布提取中，面向对象方法有助于提高水稻种植分布提取精度，且RF算法优于SVM算法。其中SNIC图像分割结合RF模型具有最高提取精度，总体精度和Kappa系数分别为96.83%、0.934，经实测数据验证，水稻实测精度为95.43%，可满足区域水稻种植分布和面积监测需求。 |
| 关 键 词： | 遥感 物候学 面向对象法 简单非迭代聚类 灰度共生矩阵 Google Earth Engine（GEE） |
| 收稿时间： | 2022-02-21 |
| 修稿时间： | 2022-06-10 |
Object-oriented extraction of paddy rice planting areas using phenological features from the GEE platform |
|
| Liu Tong, Ren Hongrui. Object-oriented extraction of paddy rice planting areas using phenological features from the GEE platform[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2022, 38(12): 189-196. DOI: 10.11975/j.issn.1002-6819.2022.12.022 | |
| Authors: | Liu Tong Ren Hongrui |
| Affiliation: | 1.Department of Geomatics, Taiyuan University of Technology, Taiyuan 030024,China |
| Abstract: | Abstract: The planting distribution of rice has been widely extracted to identify the flooding characteristics of paddy rice during the transplanting period. Optical remote sensing images can be mostly used as the data source at present. However, the transplanting period cannot characterize full spectral characteristics of the paddy rice lifespanthe, spectral characteristics of paddy rice should be considered in multiple phenological periods. The pixel-based extraction of paddy rice planting distribution can also be susceptible to the data source noise. Fortunately, an object-oriented extraction can be selected to effectively reduce the impact of data source noise in the field. Since the remote sensing images are limited by the acquisition and processing costs, only a few studies focused on the application of multiple paddy rice phenological stages in the extraction of paddy rice planting distribution in large areas. Alternatively, the emergence of the Google Earth Engine (GEE) platform in earth science data and analysis can be accessible to large amounts of remote sensing data for free and with high efficiency. Taking Panjin City, Liaoning Province of China as the research area, this study aims to realize the object-oriented extraction of rice planting areas using phenological features from the GEE platform. Four phenological periods of paddy rice were selected, namely the sowing, transplanting, heading, and maturity period. Specifically, the sowing period with the Bare Soil Index (BSI) was selected from March 15 to April 30. The transplanting period was selected as the Green Chlorophyll Vegetation Index (GCVI) and the Modified normalized difference water index (MNDWI) from May 10 to June 20. The heading period was the Normalized Difference Red Edge Index (NDRE) and Normalized difference vegetation index (NDVI) from June 30 to September 10. The maturity period with the Plant Senescence Reflectance Index (PSRI) was selected from September 20 to October 20. The 2020 Sentinel-2 time series images were filtered to construct the datasets in the four paddy rice phenological periods. Then, the spectral indices corresponding to each phenological period were calculated and synthesized by the median. Finally, the six images were synthesized into one multi-band image as the original image. The images were segmented with the Simple Non-Iterative Clustering (SNIC) available in the GEE platform. Texture features were also calculated with the Gray Level Co-occurrence Matrix (GLCM). Six models were also established for the paddy rice distribution using the random forest (RF) and Support Vector Machine (SVM), including the pixel-based RF model (PB_RF), object-oriented RF model (SNIC_RF), object-oriented RF model with texture features (SNIC_GLCM_RF), pixel-based support SVM model (PB_SVM), object-oriented SVM model (SNIC_SVM), and object-oriented SVM model with texture features (SNIC_GLCM_SVM). A full extraction of paddy rice planting distribution was implemented using the six models in the study area in 2020. An optimal model was achieved to verify the model accuracy in the field survey of paddy rice. The results show that the RF performed better than the SVM in the extraction of paddy rice planting distribution. The object-oriented method can be widely expected to improve the extraction accuracy of paddy rice planting distribution. The highest extraction accuracy was also achieved in the SNIC_RF model. Correspondingly, the overall accuracy and Kappa coefficient were 96.83% and 0.934, respectively. The field survey data was used to verify the model, where the field survey accuracy of paddy rice was 95.43%, fully meeting the requirements of regional paddy rice planting distribution and area monitoring. |
| Keywords: | remote sensing phenology object-oriented method simple non-iterative clustering gray level co-occurrence matrix Google Earth Engine |
| 点击此处可从《农业工程学报》浏览原始摘要信息 | |
| 点击此处可从《农业工程学报》下载全文 | |