| 引用本文: | 程嘉欣,李远洋,杜 斌,等.基于Sentinel-2和LSTM模型的菜心种植分布识别[J].灌溉排水学报,2026,45(10):42-49. |
| Cheng Jiaxin,Li Yuanyang,Du Bin,et al.基于Sentinel-2和LSTM模型的菜心种植分布识别[J].灌溉排水学报,2026,45(10):42-49. |
|
| 摘要: |
| 【目的】利用Sentinel-2多时序影像与双层长短期记忆网络（LSTM）模型，识别汉延渠灌区菜心空间分布。【方法】针对菜心种植地块分散、传统遥感识别方法精度较低的问题，以Sentinel-2 L2A时序影像为主要数据源，结合野外实地调研数据，构建了2025年3—12月覆盖菜心全生育期的多特征时序数据集，分析菜心与春小麦、玉米、水稻等主要作物在归一化植被指数（NDVI）时序曲线上的特征差异。利用双层LSTM算法，构建了菜心种植分布识别模型，进一步分析了模型分类误差的来源及其空间分布。【结果】菜心NDVI序列呈典型的短周期、高频“峰-谷-峰”交替特征，其NDVI峰值范围为0.7~0.9、谷值范围为0.1~0.2，与单季作物的单峰模式差异显著；所构建的双层LSTM模型总体精度达97.39%，Kappa系数为0.92，其中，菜心种植分布的准确度为0.93，非菜心准确度为0.98。【结论】所构建的识别算法能够有效提取菜心种植分布，可为灌区尺度水资源和农业管理提供参考。 |
| 关键词: 时序深度学习 卫星遥感 植被指数 汉延渠灌区 菜心 |
| DOI：10.13522/j.cnki.ggps.2026075 |
| 分类号: |
| 基金项目: |
|
| Mapping the spatial distribution of choi sum using Sentinel-2 imagery and an LSTM model |
|
Cheng Jiaxin, Li Yuanyang, Du Bin, Ye Yili, Zhang Jia, Ding Yimin
|
|
1. School of Civil and Hydraulic Engineering, Ningxia University, Yinchuan 750021, China;
2. Ningxia Key Laboratory of Digital Water Regulation of the Yellow River Water Network, Yinchuan 750021, China;
3. Zhongshui Northern Survey, Design and Research Co., Ltd., Tianjin 300222, China;
4. Ningxia Hydraulic Research Institute, Yinchuan 750021, China
|
| Abstract: |
| 【Objective】Choi sum is a vegetable often grown in fragmented smallholder plots in China. Accurately mapping its regional planting distribution is important for agricultural management but remains challenging. This study developed a remote-sensing-based method to address this challenge.【Method】The method was based on multi-temporal Sentinel-2 imagery and a two-layer, long short-term memory (LSTM) model. Sentinel-2 L2A time-series imagery, combined with field survey data, was used as the primary data source. A multi-feature time-series dataset covering the entire growth period of choy sum from March to December 2025 was constructed. The differences in time-series patterns of the normalized difference vegetation index (NDVI) between choy sum and other crops, including spring wheat, maize and rice, were analyzed. Based on these temporal differences, a two-layer LSTM-based model was developed to identify the spatial distribution of choy sum.【Result】Temporally, the time series of the NDVI of choy sum exhibited alternating peak-trough fluctuations, with peak values ranging from 0.7 to 0.9 and trough values ranging from 0.1 to 0.2. Such a pattern differed markedly from the single-peak pattern of single-season crops. Overall, the LSTM model achieved an accuracy of 97.39% for all crops, with a Kappa coefficient of 0.92. Its accuracy for choy sum and other crops was 0.93 and 0.98, respectively.【Conclusion】The proposed two-layer LSTM-based method can effectively map the spatial distribution of choy sum. It provides a useful approach for assessing spatial planting patterns and supporting water-resource management and precision agriculture for irrigation districts. |
| Key words: time-series deep learning satellite remote sensing vegetation index hanyanqu irrigation district flowering Chinese cabbage |
