EN中文

时空信息学报 >

2025 , Vol. 32 >Issue 04: 376 - 385

DOI: https://doi.org/10.20117/j.jsti.202504011

时空感知

基于方向自适应与多尺度感知的丘陵烟田遥感识别方法

  • LI Wei ,
  • QU Yani ,
  • GAO Xiang ,
  • YIN Huan ,
  • PENG Shuguang ,
  • ZHAO Huige ,
  • CHEN ,
  • Jie
展开
  • 1.湖南省烟草公司,长沙 410004;
    2.中南大学 地球科学与信息物理学院,长沙 410083;
    3.长沙市天心阁大数据研究院,长沙 410012
李伟,研究方向为烟叶生产。E-mail:343801680@qq.com
陈杰,研究方向为遥感影像智能分析与理解研究。E-mail:cj2011@csu.edu.cn

收稿日期: 2025-05-21

  修回日期: 2025-07-16

  网络出版日期: 2026-05-06

基金资助

湖南省自然科学基金项目（2023JJ30655）

收起

Remote sensing recognition of hilly tobacco fields based on directional adaptation and multiscale perception

  • 李伟 ,
  • 屈亚妮 ,
  • 高翔 ,
  • 殷欢 ,
  • 彭曙光 ,
  • 赵辉革 ,
  • 陈杰
Expand
  • 1. Hunan Provincial Tobacco Company, Changsha, 410004, China;
    2. School of Geosciences and Info-Physics, Central South University, Changsha 410083, China;
    3. Tianxin Pavilion Big Data Institute, Changsha 410012, China

Received date: 2025-05-21

  Revised date: 2025-07-16

  Online published: 2026-05-06

Supported by

Natural Science Foundation of Hunan Province (2023JJ30655)

Fold

摘要

烟田监测是实现烟草种植精细化管理的关键环节,对于优化耕作措施、提升产量与质量,以及制定精准 扶持政策具有重要意义。然而,当前在丘陵地区开展烟田遥感识别仍面临如下两大主要挑战：一是烟田块形状不 规则、尺度多样导致的有效特征提取困难;二是地形起伏和植被类型多样造成的背景干扰,使得难以实现精细化 识别。因此,本文提出一种基于方向自适应与多尺度感知（directional adaptation and multiscale perception,DAMP） 的丘陵烟田遥感智能识别方法。首先,根据丘陵烟田形状多变的特点,提出方向自适应注意力机制,通过结合不 同方向上的平均池化、最大池化和特征加权,有效捕获田块局部细节中关键低频信息（如色调信息）与高频信息（如边界信息）;其次,设计基于特征金字塔的多尺度特征增强模块,用于优化多尺度特征表达,以应对烟田尺 度多样的问题;最后,针对丘陵烟田分布零散和种植背景复杂的问题,引入 Swin Transformer 捕获全局上下文信 息,以显著提升烟田识别精度。结果表明,本文方法通过结合局部细节、多尺度特征与全局上下文信息,显著增 强了对丘陵烟田区域与复杂背景的区分能力;相较于基准模型、FCN、DeepLabV3+等已有主流模型,本文方法 精度表现最优,丘陵烟田提取的平均交并比、总体精度、F1 分数,分别达到了 76.68%、96.18%、81.76%。

关键词： 烟田识别; 方向自适应; 多尺度感知; 全局上下文; 深度学习

本文引用格式

LI Wei , QU Yani , GAO Xiang , YIN Huan , PENG Shuguang , ZHAO Huige , CHEN , Jie . 基于方向自适应与多尺度感知的丘陵烟田遥感识别方法[J]. 时空信息学报, 2025 , 32(04) : 376 -385 . DOI: 10.20117/j.jsti.202504011

Abstract

[Objective] Tobacco field monitoring is a critical component of precision management in tobacco cultivation, vital for optimizing farming practices, enhancing yield and quality, and informing targeted support policies. However, the remote sensing-based recognition of tobacco fields in hilly and mountainous regions presents significant challenges stemming from irregular plot shapes, diverse scales, considerable topographic variation, and heterogeneous vegetation backgrounds. These factors collectively impede effective feature extraction and diminish the accuracy of tobacco field identification. This study introduces an intelligent recognition method, named directional adaptation and multiscale perception (DAMP), designed to surmount these challenges and improve the precision of tobacco field extraction within complex hilly environments.
[Method] The proposed DAMP method tackles the challenges inherent in identifying hilly tobacco fields by integrating three core components: a direction-adaptive attention mechanism, a multiscale feature enhancement module, and global context modeling. First, the direction-adaptive attention mechanism synergistically combines directional average pooling, max pooling, and feature weighting to effectively capture both low-frequency attributes—such as color tone—and high-frequency details, including field boundaries. This enhances the extraction of textures characteristic of irregular plots. Second, to accommodate the varied scales of tobacco fields, a feature pyramid network-based multiscale enhancement module is employed to ensure robust feature representation across multiple resolutions. Third, to mitigate interference from complex vegetation and terrain, the Swin Transformer is incorporated to model long-range dependencies and provide global contextual information. This integration improves semantic understanding and bolsters overall recognition accuracy.
[Result] Experiments were conducted using remote sensing datasets sourced from representative tobacco-growing regions characterized by hilly terrain. The proposed DAMP approach achieved a mean intersection over union (mIoU) of 76.68%, an overall accuracy (OA) of 96.18%, and an F1 score of 81.76%. Comparative evaluations reveal that DAMP significantly outperforms conventional convolution- based models (e.g., FCN, U-Net, and DeepLabV3+) as well as the pure transformer-based Swin Transformer model. The synergistic combination of local detail extraction, multiscale feature fusion, and global context modeling proved particularly effective for discriminating tobacco fields from complex background environments.
[Conclusion] The DAMP method offers a novel and effective solution to the persistent challenge of tobacco field recognition in difficult hilly and mountainous areas. By integrating direction-aware attention mechanisms, multiscale feature pyramids, and transformer-based global context perception, the method substantially enhances both the discriminative capability and robustness of tobacco field extraction. These findings indicate that DAMP provides a promising framework for fine-scale agricultural remote sensing applications and holds potential for extension to other crops and landscapes exhibiting similar structural complexities.

Key words： tobacco field recognition; direction adaptation; multiscale awareness; global context; deep learning

参考文献

陈妮, 应丰, 王静, 李健.2021.基于 U-Net 的高分辨率遥感图像土地利用信息提取.遥感技术与应用, 36(2):285-292[Chen N,Ying F, Wang J, Li J.2021.Research on land use information extraction based on U-Net.Remote Sensing Technology and Application, 36(2):285-292 (in Chinese)]
陈仲新, 任建强, 唐华俊, 史云, 冷佩, 刘佳, 王利民, 吴文斌, 姚艳敏, 哈斯图亚.2016.农业遥感研究应用进展与展望.遥感学报, 20(5):748-767[Chen Z X, Ren J Q, Tang H J, Shi Y, Leng P,Liu J, Wang L M, Wu W B, Yao Y M, Ha S T Y.2016.Progress and perspectives on agricultural remote sensing research and applications in China.Journal of Remote Sensing, 20(5):748-767 (in Chinese)]
董梅, 苏建东, 刘广玉, 杨举田, 陈秀斋, 田雷, 王梅勋.2014.面向对象的无人机遥感影像烟草种植面提取和监测.测绘科学,39(9):87-90[Dong M, Su J D, Liu G Y, Yang J T, Chen X Z, Tian L, Wang M X.2014.Extraction of tobacco planting areas from UAV remote sensing imagery by object-oriented classification method.Science of Surveying and Mapping, 39(9):87-90 (in Chinese)]
付必环.2022.基于深度语义分割的无人机影像烟草提取方法研究.硕士学位论文.昆明:昆明理工大学[Fu B H.2022.Research on tobacco extraction method from UAV images based on deep semantic segmentation.Master Dissertation.Kunming:Kunming University of Science and Technology (in Chinese)]
赫晓慧, 吴凯旋, 李盼乐, 乔梦佳, 程淅杰.2025.基于大幅面 SAM的遥感影像建筑物提取研究.时空信息学报, 32(2):148-157[He X H, Wu K X, Li P L, Qiao M J, Cheng X J.2025.Building extraction from remote sensing images based on a large-scale SAM model.Journal of Spatio-temporal Information, 32(2):148-157 (in Chinese)]
刘诗雨, 郑文武.2024.基于改进 DeepLabV3+的涉县石堰梯田遥感提取方法研究.时空信息学报, 31(6):722-731[Liu S Y, Zheng W W.2024.A method for extraction of Shiyan terraced fields in Shexian County from remote sensing images based on improved DeepLabV3+ network.Journal of Spatio-temporal Information, 31(6):722-731 (in Chinese)]
刘秀慧, 李勇, 葛莹, 王鸿燕, 赖美芸, 顾祯蓉, 储思敏, 丁涵.2025.一种改进 U-Net 的干旱区地表水资源遥感提取研究.时空信息学报, 32(2):158-167[Liu X H, Li Y, Ge Y, Wang H Y, Lai M Y,Gu Z R, Chu S M, Ding H.2025.An improved U-Net for extraction of surface water resources from remote sensing images in arid regions.Journal of Spatio-temporal Information, 32(2):158-167 (in Chinese)]
刘宇承, 罗芳, 杜清运, 黄文丽, 石仪玮.2023.深圳市宝安区土地利用遥感分类方法对比分析.时空信息学报, 30(4):491-499[Liu Y C, Luo F, Du Q Y, Huang W L, Shi Y W.2023.Comparative analysis of remote sensing classification methods for land use in Bao'an district, Shenzhen.Journal of Spatio-temporal Information, 30(4):491-499 (in Chinese)]
潘琛, 顾建祥, 岳照溪.2024.基于高光谱影像的长三角一体化示范区典型区域植被类型分类.时空信息学报, 31(6):698-709[Pan C, Gu J X, Yue Z X.2024.Vegetation classification in typical areas of the demonstration zone of integrated regional development of Yangtze River Delta based on hyperspectral satellite images.Journal of Spatio-temporal Information, 31(6):698-709 (in Chinese)]
陶健, 申国明, 徐宜民, 梁洪波.2015.遥感在烟草种植监测及管理中的应用展望.中国烟草学报, 21(2):111-116[Tao J, Shen G M, Xu Y M, Liang H B.2015.Prospect of applying remote sensing to tobacco planting monitoring and management.Acta Tabacaria Sinica, 21(2):111-116 (in Chinese)]
薛宇飞, 张军, 张萍, 李宇宸.2022.基于 Sentinel-2 遥感影像的烟草种植信息精准提取.中国烟草科学, 43(1):96-106[Xue Y F, Zhang J, Zhang P, Li Y C.2022.Object-oriented accurate extraction of tobacco information based on sentinel-2 remote sensing images.Chinese Tobacco Science, 43(1):96-106 (in Chinese)]
张雷.2022.基于深度学习的山区烟草种植面积提取及株数统计研究.硕士学位论文.焦作:河南理工大学[Zhang L.2022.Deep learning-based research on extraction of tobacco planting area and plant counting in mountainous areas.Master Dissertation.Jiaozuo:Henan University of Science and Technology (in Chinese)]
张阳, 屠乃美, 陈舜尧, 谢会雅, 傅雪平, 邓浏平.2020.基于 Sentinel-2A 数据的县域烤烟种植面积提取分析.烟草科技, 53(11):15-22[Zhang Y, Tu N M, Chen S Y, Xie H Y, Fu X P, Deng L P.2020.Extraction of flue-cured tobacco planting area on county scale based on Sentinel-2A data.Tobacco Science & Technology, 53(11):15-22 (in Chinese)]
Ai H F, Zhu X M, Han Y Q, Ma S N, Wang Y A, Ma Y H, Qin C, Han X Y, Yang Y X, Zhang X L.2025.Extraction of levees from paddy fields based on the SE-CBAM UNet model and remote sensing images.Remote Sensing, 17(11):1871
De Macedo M M G, Mattos A B, Oliveira D A B.2020.Generalization of convolutional LSTM models for crop area estimation.IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 13:1134-1142
Du Z R, Yang J Y, Ou C, Zhang T T.2019.Smallholder crop area mapped with a semantic segmentation deep learning method.Remote Sensing, 11(7):888
Dubinin N E.2016.The SW-MSA calculation of self-diffusion coefficients in liquid lithium and rubidium.Acta Physica Polonica A, 129(3):310-312
Fan L L, Zhou Y, Liu H M, Li Y J, Cao D P.2023.Combining Swin Transformer with UNet for remote sensing image semantic segmentation.IEEE Transactions on Geoscience and Remote Sensing, 61:5530111
Han K, Wang Y H, Chen H T, Chen X H, Guo J Y, Liu Z H, Tang Y H, Xiao A, Xu C J, Xu Y X, Yang Z H, Zhang Y M, Tao D C.2023.A survey on vision transformer.IEEE Transactions on Pattern Analysis and Machine Intelligence, 45(1):87-110
He L, Liao K W, Li Y X, Li B, Zhang J L, Wang Y, Lu L M, Jian S C, Qin R, Fu X J.2024.Extraction of tobacco planting information based on UAV high-resolution remote sensing images.Remote Sensing, 16(2):359
Hou Q B, Zhou D Q, Feng J S.2021.Coordinate attention for efficient mobile network design//2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 13708-13717
Hu J, Shen L, Sun G.2018.Squeeze-and-excitation networks//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition,7132-7141
Huang L, Wu X Q, Peng Q Z, Yu X Q.2021.Depth semantic segmentation of tobacco planting areas from unmanned aerial vehicle remote sensing images in plateau mountains.Journal of Spectroscopy, 2021(1):6687799
Li T, Wang H J, Li G Z, Liu S S, Tang L.2022.SwinF:Swin Transformer with feature fusion in target detection.Journal of Physics:Conference Series, 2284(1):012027
Li W J, Fu H H, Yu L.2017.Deep convolutional neural network based large-scale oil palm tree detection for high-resolution remote sensing images//2017 IEEE International Geoscience and Remote Sensing Symposium (IGARSS),846-849
Lin T Y, Dollár P, Girshick R, He K M, Hariharan B, Belongie S.2017.Feature pyramid networks for object detection//2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR),936-944
Liu M, Fu B L, Xie S Y, He H C, Lan F W, Li Y Y, Lou P Q, Fan D L.2021a.Comparison of multi-source satellite images for classifying marsh vegetation using DeepLabV3 Plus deep learning algorithm.Ecological Indicators, 125:107562
Liu Z, Lin Y T, Cao Y, Hu H, Wei Y X, Zhang Z, Lin S, Guo B N.2021b.Swin Transformer:Hierarchical vision transformer using shifted windows//2021 IEEE/CVF International Conference on Computer Vision (ICCV),9992-10002
Shao Y F, Pan P, Zhao H X, Li J L, Yu G P, Zhou G M, Zhang J H.2025.Semantic segmentation of rice fields in sub-meter satellite imagery using an HRNet-CA-enhanced DeepLabV3+ framework.Remote Sensing, 17(14):2404
Siddique N, Paheding S, Elkin C P, Devabhaktuni V.2021.U-Net and its variants for medical image segmentation:A review of theory and applications.IEEE Access, 9:82031-82057
Villa M, Dardenne G, Nasan M, Letissier H, Hamitouche C, Stindel E.2018.FCN-based approach for the automatic segmentation of bone surfaces in ultrasound images.International Journal of Computer Assisted Radiology and Surgery, 13(11):1707-1716
Xu Y, Xue X Y, Sun Z, Gu W, Cui L F, Jin Y K, Lan Y B.2023.Deriving agricultural field boundaries for crop management from satellite images using semantic feature pyramid network.Remote Sensing, 15(11):2937
Options
文章导航

/

〈 〉