| 引用本文: | 余午阳,王一博,陈新国,等.基于阈值特征-机器学习的青铜峡灌区多年种植结构识别[J].灌溉排水学报,2023,42(12):28-35. |
| YU Wuyang,WANG Yibo,CHEN Xinguo,et al.基于阈值特征-机器学习的青铜峡灌区多年种植结构识别[J].灌溉排水学报,2023,42(12):28-35. |
|
| 摘要: |
| 【目的】基于阈值特征-随机森林算法对青铜峡灌区多年种植结构进行识别。【方法】以青铜峡灌区为研究对象，在实地调研和目视解译的基础上，基于谷歌地球引擎（Google Earth Engine）平台，采用阈值特征-随机森林方法识别2013—2020年青铜峡灌区主要粮食作物（春小麦、春玉米、水稻）的种植结构。【结果】阈值特征-随机森林算法能够用于干旱灌区多年种植结构识别，总体分类精度为0.88，Kappa系数为0.76。春小麦、春玉米、水稻的遥感提取面积与统计种植面积之间的线性拟合决定系数（R2）分别为0.80、0.93和0.86；2013—2020年作物种植面积由大到小分别为春玉米>水稻>春小麦，春玉米种植面积呈持续上升趋势，春小麦种植面积呈先升高后下降的变化趋势，水稻种植面积呈下降趋势；春玉米主要集中在灌区的北部和南部，种植区域呈南移趋势；春小麦主要集中在灌区的中部和南部，水稻主要集中在灌区中部地区，种植区域呈向北迁移的趋势。【结论】阈值特征-随机森林算法能够较好地适用于干旱灌区多年种植结构的识别，为长时间序列种植结构识别提供了新方法和思路。 |
| 关键词: 种植结构；特征指数；随机森林；青铜峡灌区；粮食作物 |
| DOI：10.13522/j.cnki.ggps.2023251 |
| 分类号: |
| 基金项目: |
|
| Extracting Crop Structure from Remote Sensing Images of Qingtongxia Irrigation District Using the Threshold Feature-machine Learning Method |
|
YU Wuyang, WANG Yibo, CHEN Xinguo, HUANG Quanzhong, HUANG Guanhua
|
|
1. College of Water Resources and Civil Engineering, China Agricultural University, Beijing 100083, China;
2. Center for Agricultural Water Research in China, China Agricultural University, Beijing 100083, China;
3. Chinese-Israeli International Center for Research and Training in Agriculture, Beijing 100083, China
|
| Abstract: |
| 【Objective】Tracking the change in cultivation areas of different crops in a basin or catchment is essential for agricultural management but challenging. The objective of this paper is to investigate the feasibility of threshold feature - random forest classification model for extracting crop structure from remote sensing imageries of irrigation district. We took Qingtongxia irrigation as an example.【Method】Using field survey and visual interpretation, the threshold feature-random forest classification method in the Google Earth Engine platform was used to delineate the planting areas of spring wheat, spring maize, and rice within the Qingtongxia irrigation district from 2013 to 2020.【Result】The threshold feature-random forest classification method is effective for extracting planting structure in the Qingtongxia irrigation district in the seven years, with an overall accuracy of 0.88 and a Kappa coefficient of 0.76. The linear fitting between the extracted areas from the remote sense images and areas obtained from survey gave a R2 which was 0.80, 0.93 and 0.86, for spring wheat, spring maize and rice, respectively. Data analysis revealed that the planting area of spring maize surpassed that of rice and spring wheat in 2013 to 2020. Spring maize cultivation had notably increased, while spring wheat cultivation initially increased and then declined; rice cultivation also decreased. Spring maize was mainly grown in the North and South of the irrigation district, while the overall irrigation area increased from the North to the South. Spring wheat was primarily cultivated in central and Southern parts of the irrigation district, and rice was in the central region. Spring wheat and rice cultivation areas increased from the South to the North.【Conclusion】The threshold feature-random forest classification model is effective and accurate for extracting annual change in planting structures in irrigation districts in arid regions. It is novel and perspective for understanding long-term evolution in planting structure in a region. |
| Key words: crop structure; characteristic index; random forest; Qingtongxia irrigation district; grain crop |
