引用本文:
【打印本页】   【下载PDF全文】   【查看/发表评论】  【下载PDF阅读器】  【关闭】
←前一篇|后一篇→ 过刊浏览    高级检索
本文已被：浏览 59次   下载 69次 本文二维码信息
码上扫一扫！
分享到： 微信 更多
字体:加大+|默认|缩小-
基于 SAM 特征编码的新造油茶林提取方法分析
邓洪涛,薛冬冬,问恪尧,杨佐兵,具琳静,刘梦曦
1.广东省岭南院勘察设计有限公司;2.中山大学 地理科学与规划学院;3.广东省林业调查规划院
摘要:
为了提高深度学习模型对油茶林的提取精度, 研究基于目视标注和筛查, 构造了一个新造油 茶林分布数据集, 并提出了一种基于 SAM (Segment anything model) 特征编码的新造油茶林提取网络 CameFormer。 该网络以 MaskFormer 模型为基础架构, 整合了 SAM 分割大模型的特征编码能力, 联合提 取像素级和实例级特征, 用于从高分辨率遥感影像中识别新造油茶林。 实验结果显示, CameFormer 模型 的 F1 分数和 IoU 值分别达到 80. 05%和 66. 73%, 明显优于 UNet 等经典模型。 CameFormer 能在 0. 5 m 分 辨率遥感影像上高精度稳定检测具有明显新造纹理特征的油茶林。
关键词:  油茶  深度学习  SAM  高分辨率影像
DOI：10. 20221 / j. cnki. 2096-2053. 202504018
分类号:
基金项目:国家自然科学基金青年科学基金项目 (C 类) (42501428), 中国博士后科学基金面上项目 (2024M763736), 广东省油茶 产业资源普查项目 (LC-2024110)。
Analysis of Extraction Network for Newly Planted Camellia oleifera Forests Based on SAM Feature Encoding
denghongtao1, Dongdong Xue2, Keyao Wen3, Zuobing Yang4, Linjing Ju2, Mengxi Liu3
1.(Guangdong Lingnan Institute Survey and Design Co. , Ltd,;2.Guangdong Lingnan Institute Survey and Design Co. , Ltd;3.School of Geography and Planning, Sun Yat-sen University;4.Guangdong Forestry Survey and Planning Institute,
Abstract:
To enhance the extraction accuracy of deep learning models for Camellia oleifera forests, this study constructed a dataset of newly planted C. oleifera forests through visual annotation and screening. A dedicated extraction network named CameFormer was proposed, integrating SAM ( Segment anything model) for feature encoding. Based on the MaskFormer architecture, this network jointly extracts pixel-level and instancelevel features to identify newly planted C. oleifera forests from high-resolution remote sensing images. The experimental results demonstrate that CameFormer achieves F1 -scores and IoU values of 80. 05% and 66. 73%, respectively, significantly outperforming classical models such as UNet. The proposed model enables high-precision and stable detection of newly planted C. oleifera forests with distinct textural characteristics on 0. 5 m resolution remote sensing imagery.
Key words:  Camellia oleifera  deep learning  SAM  high-resolution image