Abstract: The accuracy of land use data significantly impacts the simulation study of stormwater risks. Complex nonlinear relationships exist among different land features in land use classification. In order to enhance the accuracy of land classification data, this research introduces the Backpropagation(BP)neural network model with nonlinear mapping capabilities. A remote sensing image land use classification method based on deep learning is proposed. The study utilizes GF-2 remote sensing image data from the Yesanpo Scenic Area to conduct multiscale segmentation on the image. Spectral data reflecting land use information, along with DEM data and slope data, are chosen as input layer neurons, while land use types - DOI:10.3969/j.issn.1000-1565.2024.02.012基于BP神经网络的土地利用智能分类识别与雨洪风险模拟姜艳波¹,徐宁伟²,陈泰熙³,秦安臣¹,黄大庄¹(1.河北农业大学 园林与旅游学院,河北 保定 071001;2.北京城市学院 城市建设学部,北京 100191;3.香港浸会大学 计算机科学系,香港 999077 )摘 要:土地利用分类数据的精度对雨洪风险淹没模拟研究具有重要影响.土地利用分类中不同地物之间存在复杂的非线性关系,为提高土地分类数据的精度,本研究引入具有非线性映射能力的BP神经网络模型,提出了一种基于深度学习的遥感影像土地利用分类方法.选取野三坡风景名胜区GF-2遥感影像数据,对该影像进行多尺度分割.同时将能够反映土地利用信息的光谱数据和DEM数据、坡度数据,作为输入层神经元,将土地利用类型作为输出层神经元,归一化处理后进行迭代训练,构建了基于BP神经网络的土地利用分类模型.该模型的分类总体精度达到91%,Kappa系数为0.890 6.基于该模型的识别结果,利用水文模型和ArcGIS空间分析工具,模拟并分析野三坡景区百年一遇的极端降水事件造成的雨洪淹没区,并提出应对雨洪灾害的相关策略.关键词:BP神经网络;土地利用分类;机器学习;雨洪风险;野三坡风景名胜区中图分类号:TP391 文献标志码:A 文章编号:1000-1565(2024)02-0208-08Intelligent classification of land use based on BP neural network and stormwater simulationJIANG Yanbo¹, XU Ningwei², CHEN Taixi³, QIN Anchen¹, HUANG Dazhuang¹(1. College of Landscape Architecture and Tourism, Hebei Agricultural University, Baoding 071001, China; 2. Urban Construction Department, Beijing City University, Beijing 100191, China; 3. Department of Computer Science, Hong Kong Baptist University, Hong Kong 999077, China)Abstract:The accuracy of land use data significantly impacts the simulation study of stormwater risks. Complex nonlinear relationships exist among different land features in land use classification. In order to enhance the accuracy of land classification data, this research introduces the Backpropagation(BP)neural network model with nonlinear mapping capabilities. A remote sensing image land use classification method based on deep learning is proposed. The study utilizes GF-2 remote sensing image data from the Yesanpo Scenic Area to conduct multiscale segmentation on the image. Spectral data reflecting land use information, along with DEM data and slope data, are chosen as input layer neurons, while land use types - 收稿日期:2023-10-08;修回日期:2024-01-16 基金项目:河北省文化艺术科学规划和旅游研究项目(HB22-YB026);绍兴市哲学社会科学研究“十三五”规划2019年度重点课题(135408);河北省住房和城乡建设厅科研发展基金计划项目 第一作者:姜艳波(1990—),女,满族,河北农业大学在读博士,主要从事计算机在雨洪管控中的应用研究.E-mail:jiangyanbo_jyb@126.com 通信作者:秦安臣(1961—),男,河北农业大学教授,博士生导师,主要从事遥感影像在风景区中的应用研究.E-mail:759256768@qq.com黄大庄(1963—),男,河北农业大学教授,博士生导师,主要从事生态旅游规划、园林植物应用研究.E-mail:huangdazhuang@126.com第2期姜艳波等:基于BP神经网络的土地利用智能分类识别与雨洪风险模拟serve as output layer neurons. After normalization, an iterative training process is conducted to construct a BP neural network-based land use classification model. The model achieves an overall classification accuracy of 91%, with a Kappa coefficient of 0.8906. Based on the results obtained from this model, hydrological modeling and ArcGIS spatial analysis tools are employed to simulate and analyze the rain-induced flood-affected areas resulting from a once-in-a-century extreme precipitation event in the Yesanpo Scenic Area. Relevant strategies to mitigate rainfall-induced flooding are also proposed.

Key words: BP neural network, land use classification, machine learning, stormwater risk, yesanpo scenic area