面向农机轨迹行为模式识别的频域注意力和U型残差网络

doi:10.3778/j.issn.1002-8331.2505-0335

摘要/Abstract

摘要： 农业机械轨迹作业行为模式识别是一项多变量时间序列分类任务，旨在利用轨迹数据的时空特征识别农机的行为模式。针对已有方法未能从频率角度挖掘农机轨迹的全局特性以及识别精度不足的问题，提出了一种面向农机轨迹行为模式识别的频域注意力和U型残差网络FARNet。该网络包含两个不同网络分支，用于全面挖掘农机轨迹的依赖信息。其中一个分支搭载了基于频域注意力的Transformer（transformer based on frequency attention，FAT）来挖掘农机轨迹在频域空间的全局时序依赖；另一分支部署了基于正交约束的U型残差网络（U-shaped residual network based on orthogonal constraints，URNet），其以ResUnet作为骨干网络提取轨迹特征图在不同感受野的深层语义信息，探索轨迹特征间的局部空间依赖。最后设计了一种特征对齐学习模块（feature alignment learning module，FA）来融合并对齐两个分支的输出特征，全面调节农机轨迹在全局和局部不同范围下的上下文信息，提高算法的识别性能。为验证所提方法的有效性，在真实轨迹数据集上进行了实验，结果表明，所提方法相比现有的SOTA模型在水稻和小麦收割机轨迹数据集上的F1-score提高了13.94和11.47个百分点。

关键词: 农机轨迹行为模式识别, 时间序列分类, 残差神经网络, 二维特征图, 频域注意力

Abstract: Agricultural machinery trajectory operation mode identification is a multivariate time series classification task, which aims to identify the operation mode of agricultural machinery by using the spatiotemporal features of trajectory data. Aiming at the problems that the existing methods fail to mine the global characteristics of agricultural machinery trajectories from the frequency perspective and the identification accuracy is insufficient, a frequency attention and U-shaped residual network (FARNet) for agricultural machinery trajectory operation mode identification is proposed, which contains two different network branches for comprehensively mining the dependency information of agricultural machinery trajectories. One branch is equipped with Transformer based on frequency attention (FAT) to mine the global temporal correlation of agricultural machinery trajectories in the frequency domain space. The other branch deploys U-shaped residual network based on orthogonal constraints (URNet), which uses ResUnet as the backbone network to extract the deep semantic information of trajectory feature maps in different receptive fields and explore local spatial dependencies between trajectory features. Finally, a feature alignment learning module (FA) is designed to fuse and align the output features of the two branches, comprehensively adjust the contextual information of agricultural machinery trajectories in different global and local ranges, and improve the identification performance of the algorithm. To verify the effectiveness of the proposed method, experiments are carried out on a real trajectory dataset. The results show that the proposed method improves the F1-score by 13.94 and 11.47?percentage points on the paddy and wheat harvester trajectory datasets compared with the existing SOTA model.

Key words: agricultural trajectory operation mode identification, time series classification, residual neural network, two-dimensional feature map, frequency attention

曹格齐, 翟卫欣. 面向农机轨迹行为模式识别的频域注意力和U型残差网络[J]. 计算机工程与应用, 2025, 61(16): 357-369.

CAO Geqi, ZHAI Weixin. Frequency Attention and U-Shaped Residual Network for Agricultural Machinery Trajectory Operation Mode Identification[J]. Computer Engineering and Applications, 2025, 61(16): 357-369.

参考文献

[1] ZHANG F, CHEN T H, TENG S, et al. Model construction for field operation machinery selection and configuration in wheat-maize double cropping system[J]. International Journal of Agricultural and Biological Engineering, 2021, 14(3): 82-89.
[2] 刘卉, 孟志军, 王培, 等. 基于农机空间轨迹的作业面积的缓冲区算法[J]. 农业工程学报, 2015, 31(7): 180-184.
LIU H, MENG Z J, WANG P, et al. Buffer algorithms for operation area measurement based on global navigation satellite system trajectories of agricultural machinery[J]. Transactions of the Chinese Society of Agricultural Engineering, 2015, 31(7): 180-184.
[3] LU E, XU L Z, LI Y M, et al. Modeling of working environment and coverage path planning method of combine harvesters[J]. International Journal of Agricultural and Biological Engineering, 2020, 13(2): 132-137.
[4] SUN J L, WANG Z, DING S H, et al. Adaptive disturbance observer-based fixed time nonsingular terminal sliding mode control for path-tracking of unmanned agricultural tractors[J]. Biosystems Engineering, 2024, 246: 96-109.
[5] 吴才聪, 蔡亚平, 罗梦佳, 等. 基于时间窗的农机资源时空调度模型[J]. 农业机械学报, 2013, 44(5): 237-241.
WU C C, CAI Y P, LUO M J, et al. Time-windows based temporal and spatial scheduling model for agricultural machinery resources[J]. Transactions of the Chinese Society for Agricultural Machinery, 2013, 44(5): 237-241.
[6] WANG B, DU X X, WANG Y N, et al. Multi-machine collaboration realization conditions and precise and efficient production mode of intelligent agricultural machinery[J]. International Journal of Agricultural and Biological Engineering, 2024, 17(2): 27-36.
[7] POTEKO J, EDER D, NOACK P O. Identifying operation modes of agricultural vehicles based on GNSS measurements[J]. Computers and Electronics in Agriculture, 2021, 185: 106105.
[8] XIAO Y Z, MO G Z, XIONG X Y, et al. DR-XGBoost: an XGBoost model for field-road segmentation based on dual feature extraction and recursive feature elimination[J]. International Journal of Agricultural and Biological Engineering, 2023, 16(3): 169-179.
[9] CHEN Y, ZHANG X Q, WU C C, et al. Field-road trajectory segmentation for agricultural machinery based on direction distribution[J]. Computers and Electronics in Agriculture, 2021, 186: 106180.
[10] ZHAI W X, XIONG X Y, MO G Z, et al. A Bagging-SVM field-road trajectory classification model based on feature enhancement[J]. Computers and Electronics in Agriculture, 2024, 217: 108635.
[11] ZHANG X Q, CHEN Y, JIA J P, et al. Multi-view density-based field-road classification for agricultural machinery: DBSCAN and object detection[J]. Computers and Electronics in Agriculture, 2022, 200: 107263.
[12] ZHAI W X, MO G Z, XIAO Y Z, et al. GAN-BiLSTM network for field-road classification on imbalanced GNSS recordings[J]. Computers and Electronics in Agriculture, 2024, 216: 108457.
[13] CHEN Y, QUAN L, ZHANG X Q, et al. Field-road classification for GNSS recordings of agricultural machinery using pixel-level visual features[J]. Computers and Electronics in Agriculture, 2023, 210: 107937.
[14] 罗天长笑, 翟卫欣. 农机轨迹田路分类的局部方向中心性度量聚类算法[J]. 计算机工程与应用, 2024, 60(23): 303-313.
LUO T C X, ZHAI W X. Clustering algorithm with local direction centrality measurement for agricultural machinery trajectory field-road classification[J]. Computer Engineering and Applications, 2024, 60(23): 303-313.
[15] SOUZA V M A, VEIGA P S, RIBEIRO A G R. Visemble: a fast ensemble approach for time series classification with multiple visual representations[J]. Knowledge-Based Systems, 2025, 309: 112864.
[16] HOMENDA W, JASTRZ?BSKA A, PEDRYCZ W, et al. Time series classification with their image representation[J]. Neurocomputing, 2024, 573: 127214.
[17] 吕林涛, 王鹏, 李军怀, 等. 基于时间序列的趋势性分析及其预测算法研究[J]. 计算机工程与应用, 2004, 40(19): 172-174.
LV L T, WANG P, LI J H, et al. Research on the trend analysis and predictive algorithm based on time series[J]. Computer Engineering and Applications, 2004, 40(19): 172-174.
[18] 杨一鸣, 潘嵘, 潘嘉林, 等. 时间序列分类问题的算法比较[J]. 计算机学报, 2007, 30(8): 1259-1266.
YANG Y M, PAN R, PAN J L, et al. A comparative study on time series classification[J]. Chinese Journal of Computers, 2007, 30(8): 1259-1266.
[19] STRODTHOFF N, STRODTHOFF C. Detecting and interpreting myocardial infarction using fully convolutional neural networks[J]. Physiological Measurement, 2019, 40(1): 015001.
[20] COCHRANE J H. Time series for macroeconomics and finance[Z]. Chicago: University of Chicago, 1997.
[21] 刘绪颖, 卢文达, 王剑, 等. 融合多变量序列时空信息的事件早期识别方法[J]. 计算机工程与应用, 2023, 59(17): 116-122.
LIU X Y, LU W D, WANG J, et al. Early event detection based on multivariate spatial-temporal fusion[J]. Computer Engineering and Applications, 2023, 59(17): 116-122.
[22] ROY D P, YAN L. Robust Landsat-based crop time series modelling[J]. Remote Sensing of Environment, 2020, 238: 110810.
[23] 原继东, 王志海, 韩萌, 等. 基于逻辑Shapelets转换的时间序列分类算法[J]. 计算机学报, 2015, 38(7): 1448-1459.
YUAN J D, WANG Z H, HAN M, et al. A logical shapelets transformation for time series classification[J]. Chinese Journal of Computers, 2015, 38(7): 1448-1459.
[24] KARLSSON I, PAPAPETROU P, BOSTR?M H. Generalized random shapelet forests[J]. Data Mining and Knowledge Discovery, 2016, 30(5): 1053-1085.
[25] DU M S, WEI Y X, ZHENG X W, et al. Causal and local correlations based network for multivariate time series classification[J]. Neurocomputing, 2025, 634: 129884.
[26] KARIM F, MAJUMDAR S, DARABI H, et al. Multivariate LSTM-FCNs for time series classification[J]. Neural Networks, 2019, 116: 237-245.
[27] 任欢, 王旭光. 注意力机制综述[J]. 计算机应用, 2021, 41(S1): 1-6.
REN H, WANG X G. Review of attention mechanism[J]. Journal of Computer Applications, 2021, 41(S1): 1-6.
[28] 陈海涵, 吴国栋, 李景霞, 等. 基于注意力机制的深度学习推荐研究进展[J]. 计算机工程与科学, 2021, 43(2): 370-380.
CHEN H H, WU G D, LI J X, et al. Research advances on deep learning recommendation based on attention mechanism[J]. Computer Engineering & Science, 2021, 43(2): 370-380.
[29] ZHOU T, MA Z, WEN Q, et al. FEDformer: frequency enhanced decomposed transformer for long-term series forecasting[C]//Proceedings of the 39th International Conference on Machine Learning, 2022: 27268-27286.
[30] QIN Z Q, ZHANG P Y, WU F, et al. FcaNet: frequency channel attention networks[C]//Proceedings of the 2021 IEEE/CVF International Conference on Computer Vision. Piscataway: IEEE, 2021: 763-772.
[31] MA L Y, GAO T W, JIANG H T, et al. WaveIPT: joint attention and flow alignment in the wavelet domain for pose transfer[C]//Proceedings of the 2023 IEEE/CVF International Conference on Computer Vision. Piscataway: IEEE, 2023: 7181-7191.
[32] SOKOLOVA M, LAPALME G. A systematic analysis of performance measures for classification tasks[J]. Information Processing & Management, 2009, 45(4): 427-437.
[33] ZHU L, WANG X J, KE Z H, et al. BiFormer: vision transformer with bi?level routing attention[C]//Proceedings of the 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2023: 10323-10333.
[34] WU H P, XIAO B, CODELLA N, et al. CvT: introducing convolutions to vision transformers[C]//Proceedings of the 2021 IEEE/CVF International Conference on Computer Vision. Piscataway: IEEE, 2021: 22-31.
[35] HAN H Y, ZHANG M D, HOU M, et al. STGCN: a spatial-temporal aware graph learning method for POI recommendation[C]//Proceedings of the 2020 IEEE International Conference on Data Mining. Piscataway: IEEE, 2020: 1052-1057.
[36] XU K, HU W, LESKOVEC J, et al. How powerful are graph neural networks?[J]. arXiv:1810.00826, 2018.
[37] ZHANG X X, FENG Z B, MU W S. Reformer: re-parameterized kernel lightweight transformer for grape disease segmentation[J]. Expert Systems with Applications, 2025, 265: 125757.
[38] SHU S Q, YU H W, YU J X. A novel video understanding network based on PoolFormer and transformer[C]//Proceedings of the 7th International Conference on Computer Science and Application Engineering. New York: ACM, 2023: 1-5.
[39] ILBERT R, ODONNAT A, FEOFANOV V, et al. SAMformer: unlocking the potential of transformers in time series forecasting with sharpness-aware minimization and channel-wise attention[J]. arXiv:2402.10198, 2024.
[40] SHAKER A, MAAZ M, RASHEED H, et al. SwiftFormer: efficient additive attention for transformer-based real-time mobile vision applications[C]//Proceedings of the 2023 IEEE/CVF International Conference on Computer Vision. Piscataway: IEEE, 2023: 17425-17436.
[41] DU J, ZHANG S, WU G, et al. Topology adaptive graph convolutional networks[J]. arXiv:1710.10370, 2017.