摘要
针对黄河口滨海湿地复杂的地物分布与小目标分类的难点,本文提出了一种基于改进U型卷积网络(U-Net)模型的海岸带典型地物遥感自动监测方法。在经典U-Net框架基础上,堆叠卷积层以扩展感受野,并引入卷积块注意力机制(CBAM)以提高光谱与空间特征的提取能力。研究区域位于山东省黄河入海口,采用2023年哨兵-2(Sentinel-2)遥感影像数据,通过数据预处理与增强生成样本集,并进行模型训练与评价。实验结果表明,该改进模型总体分类精度(OA)达到92.73%,平均交并比(MIoU)为77.68%,相较于传统U-Net模型分别提升2%和4.3%。本研究解决了光谱相似地物的错分问题,显著提高了对小目标(如盐地碱蓬)和线性地物(如潮沟)的分类性能,保证了地物分类的完整性与连通性。本文为滨海湿地复杂生态系统的遥感分类提供了新的技术思路,同时为湿地保护与管理提供了科学依据。
This paper addressed the challenges of complex feature distribution and small target classification in the coastal wetlands of the Yellow River Estuary.It proposed a remote sensing automatic monitoring of typical coastal features based on an improved convolutional networks for biomedical image segmentation(U-Net)model.Building upon the classic U-Net framework,convolutional layers were stacked to expand the receptive field,and a convolutional block attention module(CBAM)was introduced to improve the extraction of spectral and spatial features.The study area was located at the estuary of the Yellow River in Shandong Province,and Sentinel-2 satellite imagery from 2023 was used.Data preprocessing and augmentation were performed to generate a sample set for model training and evaluation.Experimental results show that the improved model achieves overall classification accuracy(OA)of 92.73%,and the mean intersection over union(MIoU)increases to 77.68%,representing improvements of 2%and 4.3%,respectively,compared to the traditional U-Net model.The study addresses the issue of misclassification of spectrally similar features and significantly enhances the classification performance for small targets(such as Suaeda salsa)and linear features(such as tidal channels).It ensures the completeness and connectivity of feature classification.This paper provides a new technical approach for remote sensing classification of complex coastal wetland ecosystems and offers scientific evidence for wetland protection and management.
作者
姚斌
许豪刚
王溆栋
张晨航
YAO Bin;XU Haogang;WANG Xudong;ZHANG Chenhang(Zhejiang Dixin Technology Group Company Limited,Hangzhou,Zhejiang 310000,China;Zhejiang Academy of Surveying and Mapping Science and Technology,Hangzhou,Zhejiang 311100,China)
出处
《北京测绘》
2025年第11期1615-1620,共6页
Beijing Surveying and Mapping
基金
浙江省自然资源厅2022科研项目(2022-57)。
关键词
遥感分类
黄河口滨海湿地
深度学习
U型网络改进模型
卷积块注意力机制
remote sensing classification
Yellow River estuary coastal wetlands
deep learning
improved convolutional networks for biomedical image segmentation(U-Net)model
convolutional block attention mechanism
