| 基于FCDxLSTM模型的江苏省生态类型变化区域识别技术研究 |
| Research on Identification Technology of Ecological Type Change Aeras in Jiangsu Province Based on the FCDxLSTM Model |
| 投稿时间：2025-06-30 修订日期：2025-09-16 |
| DOI：10.19316/j.issn.1002-6002.2026.02.10 |
| 中文关键词: 江苏 深度学习 变化区域识别 数据集 焦点中心差分长短期记忆网络 |
| 英文关键词:Jiangsu Province deep learning change area identification dataset FCDxLSTM |
| 基金项目:江苏省生态环境监测科研基金项目(24A15) |
|
| 摘要点击次数: 0 |
| 全文下载次数: 0 |
| 中文摘要: |
| 基于2022—2023年国产高分、资源系列卫星获取的多期遥感影像数据,结合人工目视解译,构建了覆盖江苏省13个设区市、95个区县的生态类型的变化样本数据集,总样本量6.5万组,为生态类型变化检测提供了高质量训练与验证数据。研究提出焦点中心差分长短期记忆网络(FCDxLSTM),其核心架构包含3个部分:双分支骨干网络、特征增强层和跨尺度交互融合模块。基于构建的样本数据集,开展生态类型变化检测实验,并与6种当前最优方法进行对比实验。结果表明:FCDxLSTM模型在各项评价指标中均表现最优;相较于基线模型CDxLSTM,其交并比(IoU)达0.886 9,提升幅度为1.58%;定性分析显示,该模型具备更精准的边缘检测能力,能够显著抑制伪变化现象,有效提升生态类型变化检测精度。研究结果对江苏生态质量监测评价、重要生态空间人类活动监管具有重要应用意义。 |
| 英文摘要: |
| Based on multi-temporal remote sensing image data acquired by domestic GF (Gaofen) series satellites from 2022 to 2023,combined with manual visual interpretation,an ecological type change sample dataset covering 13 prefecture-level cities and 95 counties in Jiangsu Province was constructed.With a total of 65 000 sample groups,this dataset provides high-quality training and validation data for ecological type change detection.Focal Central Difference Extended LSTM (FCDxLSTM) was proposed in this study,whose core architecture consists of three components: a dual-branch backbone network,a feature enhancement layer,and a cross-scale interactive fusion module.Using the constructed sample dataset,ecological type change detection experiments were conducted and compared with six state-of-the-art (SOTA) methods.Experimental results demonstrate that the FCDxLSTM model achieves optimal performance across all evaluation metrics.Compared with the baseline model CDxLSTM,its Intersection over Union (IoU) reached 0.886 9,showing an improvement of 1.58%.Qualitative analysis indicates that the model exhibits more precise edge detection capabilities,significantly suppresses false change phenomena,and effectively enhances the accuracy of ecological type change detection.The research findings hold important application significance for ecological quality monitoring and evaluation in Jiangsu Province,as well as human activity supervision in key ecological spaces. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|
