成都平原彭州地区耕地信息的SegFormer-PAS模型提取  

SegFormer-PAS Model Extraction of Arable Land in Pengzhou Area of Chengdu Plain

作　　者：谢美琳 刘刚[1] 何敬[1,2] 闫航嘉 李典 XIE Meilin;LIU Gang;HE Jing;YAN Hangjia;LI Dian(School of Geography and Planning,Chengdu University of Technology,Chengdu,Sichuan 610059,China;Chengdu Institute of Natural Resources Survey and Utilization,Chengdu,Sichuan 610042,China)

机构地区：[1]成都理工大学地理与规划学院,四川成都610059 [2]成都市自然资源调查利用研究院,四川成都610042

出　　处：《宜宾学院学报》2024年第12期44-50,共7页Journal of Yibin University

基　　金：国家自然科学基金项目(42371418);四川省自然资源厅科研项目(KJ-2023-024);自然资源部成都平原国土生态与土地利用野外科学观测研究站开放基金项目(CDORS-2023-04)。

摘　　要：以高分二号融合的1m分辨率数据为数据源,采用改进后的SegFormer-PAS模型对成都平原彭州地区耕地进行提取.模型选取Mit-B0作为为编码器,并引入极化自注意力机制以更好地捕获全局上下文信息,解决耕地特征提取不够充分的问题,在编码器的上采样部分采用反卷积替代简单的双线性插值以减少空间细节上的损失.实验结果表明:SegFormer-PAS模型在实验区域的交并比、召回率、准确率及F1分数分别为90.18%、91.86%、90.86%、91.26%,较基准模型SegFormer-B0均有提升;且SegFormer-PAS在成都平原彭州地区的耕地提取任务的效果均优于SegFormer、U-Net、Unet++、HRNet这4种经典语义分割算法.The improved SegFormer-PAS model is used to extract the cultivated land in Pengzhou of Chengdu Plain by using the 1 m resolution data fused by Gaofen II as the data source.The model selects Mit-B0 as the encoder,and introduces the polarized self-attention(PSA)mechanism to better capture the global context information and solve the problem of insufficient extraction of cultivated land features,and adopts the transpose convolution instead of simple bilinear interpolation in the up-sampling part of the encoder to reduce the loss of spatial details.The experimental results show that the intersection-to-union(IoU),recall,accuracy and F1 score of the SegFormer-PAS model in the experimental area are 90.18%,91.86%,90.86%and 91.26%,respectively,which are improved compared with that of the benchmark model SegFormer-B0,and the results of SegFormer-PAS in the task of cropland extraction in Pengzhou of Chengdu Plain are better than the four classical semantic segmentation algorithms,namely SegFormer,UNet,Unet++and HRNet.

关 键 词：耕地提取 深度学习 SegFormer 极化自注意力机制 高分二号影像 

分 类 号：S157.4[农业科学—土壤学]