[1]张善文,许新华.基于改进视觉Mamba-UNet模型的作物害虫检测方法[J].江苏农业学报,2026,42(06):1191-1201.[doi:doi:10.3969/j.issn.1000-4440.2026.06.011]
　ZHANG Shanwen,XU Xinhua.Crop pest detection method based on improved visual Mamba-UNet model[J].,2026,42(06):1191-1201.[doi:doi:10.3969/j.issn.1000-4440.2026.06.011]
点击复制

基于改进视觉Mamba-UNet模型的作物害虫检测方法()

江苏农业学报[ISSN:1006-6977/CN:61-1281/TN]

卷:
42
期数:
2026年06期
页码:
1191-1201
栏目:
农业信息工程
出版日期:
2026-06-30

文章信息/Info

Title:
Crop pest detection method based on improved visual Mamba-UNet model
作者:
张善文1; 2; 许新华1
(1.郑州西亚斯学院工学部,河南新郑451150；2.河南省农业信息数智化工程研究中心,河南新郑451150)
Author(s):
ZHANG Shanwen1; 2; XU Xinhua1
(1.SIAS Faculty of Engineering, Xinzheng 451150, China;2.Henan Engineering Research Center for Agricultural Information Digital & Intelligent Technology, Xinzheng 451150, China)
关键词:
作物害虫; Mamba-UNet模型; 视觉状态空间模块(VSS); 空间与通道注意力
Keywords:
crop pests; Mamba-UNet model; visual state space module (VSS); spatial and channel attention
分类号:
S433
DOI:
doi:10.3969/j.issn.1000-4440.2026.06.011
文献标志码:
A
摘要:
针对田间自然环境下作物害虫种类多样、体积微小、姿态多变及背景复杂等因素导致的检测难题,以及传统U-Net模型全局特征提取能力不足、Transformer模型计算复杂度高等问题,本研究提出一种融合空间与通道注意力视觉的Mamba-UNet模型的作物害虫检测模型——SCAFAVM-UNet。该模型以U-Mamba模型架构为基础,采用编码器-解码器结构,该模型整合了主干起始模块(Stem)、残差视觉Mamba模块(RVM)和视觉状态空间模块(VSS)、空间通道注意力视觉状态空间模块(SCAVSS)、多尺度自适应特征聚合模块(MSAFA)以及交叉二维扫描机制(CS2D)。在IP102数据集上的试验结果表明,SCAFAVM-UNet模型的检测精度与Dice相似系数(DSC)分别达到76.12%和73.11%,并且其检测精度与训练效率整体优于U-Net模型、DMSU-Net模型、MSRAL-UNet模型、TransFCL模型、Swin-UNet模型、Mamba-UNet模型以及MSV-Mamba模型。在害虫目标微小、触角及足部细节丰富且与背景对比度较低的复杂场景下,该模型能够更精准地定位害虫，并保持轮廓完整性,显著减少漏检与目标残缺现象。本研究结果为智慧农业中田间害虫的精准监测与防治决策提供了理论依据。
Abstract:
Aiming at the detection challenges caused by diverse species, small size, variable postures, and complex backgrounds of crop pests in field natural environments, as well as the insufficient global feature extraction ability of the traditional U-Net model and high computational complexity of the Transformer model, this study proposed a crop pest detection model named SCAFAVM-UNet, which is a visual Mamba-UNet model integrating spatial and channel attention. Based on the U-Mamba architecture, the model adopted an encoder-decoder structure and integrated a Stem module, residual visual Mamba module (RVM), visual state space module (VSS), spatial and channel attention visual state space module (SCAVSS), multi-scale adaptive feature aggregation module (MSAFA), and a crisscross 2D scanning mechanism (CS2D). Experimental results on the IP102 dataset showed that the detection accuracy and Dice similarity coefficient (DSC) of the SCAFAVM-UNet model reached 76.12% and 73.11%, respectively, and its overall detection accuracy and training efficiency were superior to those of the U-Net, DMSU-Net, MSRAL-UNet, TransFCL, Swin-UNet, Mamba-UNet, and MSV-Mamba models. In complex scenes with tiny pest targets, rich details of antennae and feet, and low contrast with the background, the model can locate pests more accurately and maintain contour integrity, significantly reducing missed detection and target incompleteness. The results of this study provide a theoretical basis for the accurate monitoring and control decision-making of field pests in smart agriculture.

参考文献/References:

[1]CHEN H R, WEN C J, ZHANG L, et al. Pest-PVT A model for multi-class and dense pest detection and counting in field-scale environments[J]. Computers and Electronics in Agriculture,2025,230:109864.
[2]JIAO L, XIE C J, CHEN P, et al. Adaptive feature fusion pyramid network for multi-classes agricultural pest detection[J]. Computers and Electronics in Agriculture,2022,195:106827.
[3]WANG X Q, ZHANG S W, ZHANG T. Crop insect pest detection based on dilated multi-scale attention U-Net[J]. Plant Methods,2024,20(1):34-42.
[4]熊干,陈慈发,张上. QMDF-YOLO11:复杂场景下水稻害虫检测算法[J]. 计算机工程与应用,2025,61(13):113-123.
[5]崔晓辰,雷一东. 基于深度学习的林业害虫智能化检测方法研究进展[J]. 世界林业研究,2024,37(4):53-57.
[6]杨锋,姚晓通． 基于改进YOLOv8的小麦叶片病虫害检测轻量化模型[J]. 智慧农业(中英文),2024,6(1):147-157．
[7]张小艾,乔红波,席磊,等. 基于改进YOLOv8的大田农作物害虫小目标检测方法[J]. 农业工程学报,2025,41(13):225-233.
[8]李萍,刘裕,师晓丽,等. 基于多尺度残差空间注意力轻量化U-Net的农业害虫检测方法[J]. 江苏农业科学,2023,51(3):187-196.
[9]李安琦,马丽,于合龙,等. 改进的U-Net算法在遥感图像典型农作物分类研究[J]. 红外与激光工程,2022,51(9):428-434
[10]MAYYA A M, ALKAYEM N F. Triple-stage crack detection in stone masonry using YOLO-ensemble, MobileNetV2U-net, and spectral clustering[J]. Automation in Construction,2025,172:106045.
[11]LIU Y J, YU J Z, XU Y, et al. A survey on transformer context extension:approaches and evaluation[EB/OL].(2025-07-08)
[2025-06-06]. https://doi.org/10.48550/arXiv.2503.13299.
[12]SARANYA T, DEISY C, SRIDEVI S, et al. Efficient agricultural pest classification using vision Transformer with hybrid pooled multihead attention[J]. Computers in Biology and Medicine,2024,177:108584.
[13]LIU H L, ZHAN Y Z, SUN J, et al. A Transformer-based model with feature compensation and local information enhancement for end-to-end pest detection[J]. Computers and Electronics in Agriculture,2025,231:109920.
[14]CAO H, WANG Y Y, CHEN J, et al. Swin-UNet:UNet-like pure transformer for medical image segmentation[EB/OL].(2021-05-12)
[2021-05-12]. https://doi.org/10.48550/arXiv.2105.05537.
[15]MAURCIO J, DOMINGUES I, BERNARDINO J, et al. Comparing vision transformers and convolutional neural networks for image classification:a literature review[J]. Applied Sciences, 2023,13:5521-5537.
[16]LIEBER O, LENZ B, BATA H, et al. Jamba:a hybrid Transformer-Mamba language model[EB/OL]. (2024-07-03)
[2025-05-06]. https://doi.org/10.48550/arXiv.2403.19887.
[17]LIU Y, TIAN Y J, ZHAO Y Z, et al. Vmamba: visual state space model[EB/OL]. (2024-01-18)
[2024-12-29]. https://doi.org/10.48550/arXiv.2401.10166.
[18]RAHMAN M, TUTUL A, NATH A, et al. Mamba in vision:a comprehensive survey of techniques and applications[EB/OL]. (2024-10-04)
[2025-05-08]. https://doi.org/10.48550/arXiv.2410.03105.
[19]WANG Z, ZHENG J Q, ZHANG Y, et al. Mamba-UNet:U-Net-like pure visual mamba for medical image segmentation[EB/OL]. (2024-03-30)
[2025-05-06]. https://doi.org/10.48550/arXiv.2402.05079.
[20]LIU J, YANG H, ZHOU H Y, et al. Swin-UMamba:Mamba-based U-Net with ImageNet-based pretraining[EB/OL]. (2024-03-06)
[2025-06-01]. https://doi.org/10.48550/arXiv.2402.03302.
[21]LIAO W B, ZHU Y H, WANG X Y, et al. LightM-UNet:Mamba assists in lightweight u-net for medical image segmentation [EB/OL]. (2024-03-11)
[2025-06-01]. https://doi.org/10.48550/arXiv.2403.05246.
[22]WANG Q N, WANG C L, LAI Z X, et al. InsectMamba:insect pest classification with state space model[EB/OL]. (2024-04-04)
[2025-06-02]. https://doi.org/10.48550/arXiv.2404.03611.
[23]YANG X X, WANG Q, ZHANG K Q, et al. MSV-Mamba:a multiscale vision mamba network for echocardiography segmentation[EB/OL]. (2025-01-13)
[2025-06-05]. https://doi.org/10.48550/arXiv.2501.07120.

备注/Memo

备注/Memo:
收稿日期：2025-06-05基金项目：河南省科技攻关项目(252102110343、242102210021)；河南省高等学校重点科研项目(24B510016)作者简介：张善文(1965-),男,陕西西安人,博士,教授,研究方向为深度学习及其应用。(E-mail)wjdw716@163.com
更新日期/Last Update: 2026-07-15