梁爽,刘宗放,郭祥云

• 1:北京信息科技大学计算机学院
• 2:北京信息科技大学管理科学与工程学院

摘要(Abstract)：

虫害对棉花的产量和质量影响严重，虫害识别是虫害防治的重要指施。针对许多虫害外观相似，现有算法难以区分细微差异，导致分类精度较低的问题，提出了一种基于改进YOLOv9的高效棉田虫害识别模型YOLO-ILE。引入Involution神经网络算子，提升模型对小目标的识别能力；采用大型可分离核注意力（large separable kernel attention, LSKA）模块，提升识别精度并减少计算复杂度；加入指数移动平均（exponential moving average, EMA）注意力机制，提升模型对关键特征的捕获能力。研究结果表明：在构建的新疆棉田虫害数据集上，YOLO-ILE模型的mAP@0.5和mAP@0.5∶0.95指标分别达到了93.2%、75.4%，相较于YOLOv9模型，分别提升了2.6、2.3百分点；与其他YOLO系列模型相比，在棉田虫害识别上的mAP@0.5指标达到了最优。研究成果有助于推动棉田虫害防控的数字化与智能化。

关键词(KeyWords)： YOLOv9;棉田虫害识别;注意力机制;神经网络算子;深度学习

Abstract:

Keywords:

基金项目(Foundation):

作者(Author): 梁爽,刘宗放,郭祥云

DOI: 10.16508/j.cnki.11-5866/n.2025.05.011

参考文献(References)：

• [1]雒珺瑜，张帅，任相亮，等.近十年我国棉花虫害研究进展[J].棉花学报，2017,29(S1):100-112.LUO J Y, ZHANG S, REN X L, et al. Research progress of cotton insect pests in China in recent ten years[J]. Cotton Science,2017,29(S1):100-112.(in Chinese)
• [2] ZHANG M Y,LIU C F,LI Z H,et al. From convolutional networks to vision transformers:evolution of deep learning in agricultural pest and disease identification[J]. Agronomy,2025,15(5):1079.
• [3]张友为，王鑫鑫，范晓飞.基于深度学习的玉米和番茄病虫害检测技术研究进展[J].江苏农业科学，2024,52(10):10-20.ZHANG Y W,WANG X X,FAN X F. Research progress on deep learning-based detection of diseases and pests in maize and tomato[J]. Jiangsu Agricultural Sciences,2024,52(10):10-20.(in Chinese)
• [4] PATEL V K, ABHISHEK K, SELVARAJAN S. Optimized recurrent neural network-based early diagnosis of crop pest and diseases in agriculture[J]. Discover Computing, 2024,27(1):43.
• [5]黄建荣.稻纵卷叶螟和褐飞虱为害水稻的光谱监测[D].南京：南京农业大学，2013.HUANG J R. Detection of Cnaphalocrocis medinalis Güenée and Nilaparvata lugens(St?l)damage in rice using spectral data[D]. Nanjing:Nanjing Agricultural University,2013.(in Chinese)
• [6] WANG C Y,BOCHKOVSKIY A,LIAO H Y M. YOLOv7:trainable bag-of-freebies sets new state-of-the-art for realtime object detectors[C]//2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR). Los Alamitos, CA, USA:IEEE Computer Society,2023:7464-7475.
• [7]于蕾，胡聪，张奇.自动化虫害图像智能分析系统设计[J].实验技术与管理，2022,39(12):137-142.YU L,HU C,ZHANG Q. Design of automatic pest image intelligent analysis system[J]. Experimental Technology and Management,2022,39(12):137-142.(in Chinese)
• [8]王曾龙，蒋勇勇，彭海峰，等.一种基于Faster-RCNN的棉花虫害识别与统计方法[J].大众科技，2023,25(5):5-7.WANG Z L,JIANG Y Y,PENG H F,et al. A method for cotton pests identification and statistics based on Faster-RCNN[J].Popular Science&Technology,2023,25(5):5-7.(in Chinese)
• [9]巨志勇，易成，周重臣，等. YOLO-Rice：一种基于YOLOv5的水稻虫害检测[J].控制工程，2024,31(12):2196-2205.JU Z Y,YI C,ZHOU Z C,et al. YOLO-Rice:a rice pest detection based on YOLOv5[J]. Control Engineering of China,2024,31(12):2196-2205.(in Chinese)
• [10]王泰华，郭亚州，张家乐，等.基于改进YOLO v5s的水稻害虫识别研究[J].农业机械学报，2024,55(11):39-48.WANG T H,GUO Y Z,ZHANG J L,et al. Rice pest identification based on improved YOLO v5s[J]. Transactions of the Chinese Society for Agricultural Machinery,2024,55(11):39-48.(in Chinese)
• [11]杨满仙，陈燕红，李雨晴，等.新疆棉田主要昆虫图像数据集CottonInsect[J].中国科学数据（中英文网络版），2025,10(1):62-72.YANG M X,CHEN Y H,LI Y Q,et al. A dataset of common insects in Xinjiang cotton fields:CottonInsect[J]. China Scientific Data,2025,10(1):62-72.(in Chinese)
• [12]杨玉艳，肖茹.基于改进DCGAN的数据增强方法研究：以BDD100K数据集为例[J].信息与电脑，2025,37(10):37-39.YANG Y Y,XIAO R. Research on data augmentation method based on improved DCGAN:taking the BDD100K dataset as an example[J]. Information&Computer,2025,37(10):37-39.(in Chinese)
• [13]姚百蔚，邵佳慧，徐英杰，等.基于改进YOLOv9的番茄病害检测[J].计算机技术与发展，2025,35(6):56-61.YAO B W,SHAO J H,XU Y J,et al. Tomato disease detection based on improved YOLOv9[J]. Computer Technology and Development,2025,35(6):56-61.(in Chinese)
• [14]罗希，贺强，张宁轩，等.基于改进型YOLOv9的绝缘子及缺陷检测[J].机电工程技术，2024,53(10):197-202.LUO X,HE Q,ZHANG N X,et al. Insulator and defect detection based on improved YOLOv9[J]. Mechanical&Electrical Engineering Technology,2024,53(10):197-202.(in Chinese)
• [15] TIAN Y J,YE Q X,DOERMANN D. YOLOv12:attentioncentric real-time object detectors[EB/OL].(2025-02-18)[2025-06-09]. https：//doi. org/10. 48550/arXiv. 2502. 12524.
• [16]周啟雪，余映，胡家绿.利用Involution级联注意力机制的古代壁画图像修复网络[J/OL].计算机科学，2025:1-11.(2025-02-28)[2025-03-01]. https：//link. cnki. net/urlid/50.1075. TP. 20250227. 1624. 055.ZHOU Q X,YU Y,HU J L. Ancient mural image restoration network using Involution cascaded attention mechanism[J/OL].Computer Science,2025:1-11.(2025-02-28)[2025-03-01].https：//link. cnki. net/urlid/50. 1075. TP. 20250227. 1624. 055.(in Chinese)
• [17]张翔.基于深度学习的病害孢子出芽率精准检测及系统实现[D].海口：海南大学，2022.ZHANG X. Accurate detection and system implementation of disease spore germination rate based on deep learning[D].Haikou:Hainan University,2022.(in Chinese)
• [18]刘子龙，张磊.基于LSKA-Transposed与ECA融合的YOLOv5苹果目标检测算法[J/OL].控制工程，2025:1-11.(2025-03-06)[2025-06-09]. https：//doi. org/10. 14107/j. cnki. kzgc.20240944.LIU Z L,ZHANG L. YOLOv5 apple target detection algorithm based on LSKA-Transposed and ECA fusion[J/OL]. Control Engineering of China,2025:1-11.(2025-03-06)[2025-06-09]. https：//doi. org/10. 14107/j. cnki. kzgc. 20240944.(in Chinese)
• [19] ZHAO Y Z,ZHAO H D,SHI J F. YOLOv8-MAH:multiattribute recognition model for vehicles[J]. Pattern Recognition,2025,167:111849.
• [20] OUYANG D L,HE S,ZHANG G Z,et al. Efficient multi-scale attention module with cross-spatial learning[C]//2023 IEEE International Conference on Acoustics, Speech and Signal Processing(ICASSP). New York, USA:IEEE,2023:1-5.