基于注意力机制双边UNet模型的玉米种植地识别方法研究

打开文本图片集

中图法分类号：TP751 文献标识码：A 文章编号：1000-2324（2025）04-0720-11

Corn Plantation Recognition Method Based on Attention MechanismBilateralUNetModel

1²， Xiao-tao³， SHUA Qiang-qiang1，²， QI Wen-chao4

TechnologyResearchCenterinDazhou/Sichuan University

ofArtsand Sciences，Dazhou 635ooo，China

3.SchoolofEcological Tourism/Sichuan UniversityofArtsand Science，Dazhou 635oo0，China

4.Aerospace Information Research Institute/ChineseAcademy ofSciences，BeijinglOoo94，China

Abstract：Corn is oneof thethree major staple crops in China，and the accurate estimationof its plantingarea is ofcrucial significance forsafeguardingnational foodsecurity.However，traditional groundsurveymethodsareinefficientand fail to meettheneeds for theacurateandtimelyacquisitionofinformationoncorn planting areaandits spatialdistribution. Thisstudytakescornplantingareas inChangzhiCity，ShanxiProvinceastheresearchobject，constructsatime-series dataset basedoncorn growth stagesutilizing multi-temporal Sentinel-2optical images，anddevelops a bilateral UNet network model integratedwith attention The model employs lightweight MobileNetV2asthe backbonenetwork， integratesan atrous spatial pyramid pooling module and convolutional attention module to enhance feature extraction capabilities，The model employs lightweight MobileNetV2 as the backbone network，integratesanatrous spatial pyramidpoling module and convolutional atention module to enhance feature extraction capabilities，thereby improving segmentationaccuracyforsmalldatasets.Theexperimental results indicate that： （1）With the incorporationof EVI featuresfromdiffrentgrowthstages，thenetwork'srecognitionaccuracygraduallyimproves，with thehighest segmentation acuracy achieved when using composite EVI features from six growth periods.（2） The proposed method achieves 87.97% in OA， 87.35% in F1 score of，and 82.25% in IoU，with accuracy metrics and visual results outperformingsix existing"U"-shaped networks.（3）Compared with the data fromthestatistical yearbook，theerorrate of recognized corn planting areas in Changzhi City， Shanxi Province is only 6.1% ，with the overall recognition result being closely aligned.

Keywords：Loess Plateau; Sentinel-2; deep Learning; corn plantation extraction; time-series EVI

玉米作为我国三大粮食作物之一，准确及时地获取其种植面积和空间分布信息对农业结构调整和粮食安全具有重要意义[。（剩余15622字）