融合多尺度注意力特征的林业外来害虫检测方法

doi:10.11707/j.1001-7488.LYKX20240446

林业科学 ›› 2025, Vol. 61 ›› Issue (10): 154-163.doi: 10.11707/j.1001-7488.LYKX20240446

• 研究论文 • 上一篇

融合多尺度注意力特征的林业外来害虫检测方法

张长春^1,^2,³(),杨兴昌¹,王国骅¹,王冰婧⁴,李诗杰¹,陈方舟¹,葛永泰¹,石娟^5,*(),张军国^1,^2,^3,*()

1. 北京林业大学工学院　北京 100083
2. 林木资源高效利用全国重点实验室　北京 100083
3. 北京林业大学生物多样性智慧监测研究中心　北京 100083
4. 北京林业大学信息学院　北京 100083
5. 北京林业大学林学院　北京 100083

收稿日期:2024-07-16 出版日期:2025-10-25 发布日期:2025-11-05
通讯作者: 石娟,张军国 E-mail:zhangchangchun@bjfu.edu.cn;shi_juan@263.net;zhangjunguo@bjfu.edu.cn
基金资助:
中央高校基本科研业务费专项资金（QNTD202510）；中央高校优秀青年团队项目（QNTD202304）；北京市级大学生创新训练项目（S202310022154）。

A Detection Method of Alien Forest Pests with Integrating Multi-scale Attention Features

Changchun Zhang^1,^2,³(),Xingchang Yang¹,Guohua Wang¹,Bingjing Wang⁴,Shijie Li¹,Fangzhou Chen¹,Yongtai Ge¹,Juan Shi^5,*(),Junguo Zhang^1,^2,^3,*()

1. School of Technology, Beijing Forestry University　Beijing 100083
2. State Key Laboratory of Efficient Production of Forest Resources　Beijing 100083
3. Research Center for Biodiversity Intelligent Monitoring, Beijing Forestry University　Beijing 100083
4. School of Information Science and Technology, Beijing Forestry University　Beijing 100083
5. School of Forestry, Beijing Forestry University　Beijing 100083

Received:2024-07-16 Online:2025-10-25 Published:2025-11-05
Contact: Juan Shi,Junguo Zhang E-mail:zhangchangchun@bjfu.edu.cn;shi_juan@263.net;zhangjunguo@bjfu.edu.cn

摘要/Abstract

摘要：

目的: 针对林业外来害虫因目标尺度多样、栖息环境复杂、易受遮挡等因素导致的现有检测算法识别准确率低、鲁棒性差的问题，提出一种融合多尺度注意力特征的林业外来害虫检测方法（MAF-YOLO），通过增强模型对不同尺度特别是小尺度目标的特征提取能力，并优化边界框回归策略，从而显著提升在复杂自然环境下对外来害虫的检测精度和泛化能力。方法: 该方法以YOLOv5s为基线架构，其核心改进包括：1）在颈部网络中嵌入坐标注意力机制，以强化对目标区域关键特征的捕获能力并抑制背景噪声干扰；2）增设160像素×160像素的小目标检测头，构建多层级检测结构，以提升对微小目标的检测灵敏度；3）采用Focal-EIoU损失函数替代原始CIoU损失函数，缓解正负样本及难易样本不平衡的问题，优化目标的定位精度；4）引入域适应训练策略，通过在大规模通用数据集上预训练，增强模型对不同场景的泛化能力。结果: 在包含15类潜在入侵和已入侵的林业外来害虫图像数据集上，训练和评估提出的模型，改进后的YOLOv5s模型精确率和召回率较原始YOLOv5s模型分别提升3.6%和4.4%；与SSD、YOLOv7、YOLOv8模型相比，改进模型的平均识别精度分别提高2.2%、1.1%和0.3%；结合域适应后模型精确率达77.9%，较原模型提升2.1%。结论: 本研究实现了林业外来害虫的精准识别，增强了模型的准确性与鲁棒性，为有效监测林业外来害虫提供了理论依据和技术支持。

关键词: 林业外来害虫, YOLOv5s, 坐标注意力, 域适应, 图像检测

Abstract:

Objective: To address the challenges of low identification accuracy and insufficient robustness in existing detection algorithms for alien forest pests, which stem from factors such as diverse target scales, complex habitats, and frequent occlusions, this paper proposes a detection method integrating multi-scale attention features, termed MAF-YOLO. The method aims to significantly enhance the detection accuracy and generalization capability for alien pests in complex natural environments by strengthening the model’s feature extraction capability for targets of varying scales, particularly small-scale ones, and optimize the bounding box regression strategy. Method: The proposed method was based on the YOLOv5s baseline architecture, and its core modifications included: 1) A coordinate attention (CA) mechanism was embedded into the neck network to enhance the capture of key target features and suppress background noise. 2) A small target detection head with 160×160 pixel was added to construct a multi-level detection structure, thereby improving detection sensitivity for minute objects. 3) The Focal-EIoU loss function was employed to replace the original CIoU loss, in order to mitigate the imbalance between positive/negative and easy/hard samples, and refine object localization accuracy. 4) A domain adaptation training strategy was introduced to improve the model's generalization across diverse scenarios by pre-training on a large-scale general-purpose dataset. Result: The proposed model was trained and evaluated on an image dataset comprising 15 categories of potential and existing invasive forest pests. The improved YOLOv5s model, MAF-YOLO, demonstrated an increase in precision and recall by 3.6% and 4.4%, respectively, compared to the original YOLOv5s. In comparison with the SSD, YOLOv7, and YOLOv8 models, the average precision of the improved model was higher by 2.2%, 1.1%, and 0.3%, respectively. Furthermore, with the integration of domain adaptation, the model’s precision reached 77.9%, representing a 2.1% improvement over the baseline model under the same strategy. Conclusion: This study achieves precise recognition of invasive alien forest pests, enhances the model's accuracy and robustness, and provides theoretical and technical support for effective monitoring of alien forest pest.

Key words: alien forest pest, YOLOv5s, coordinate attention, domain adaptation, image detection

中图分类号:

S763.30

张长春,杨兴昌,王国骅,王冰婧,李诗杰,陈方舟,葛永泰,石娟,张军国. 融合多尺度注意力特征的林业外来害虫检测方法[J]. 林业科学, 2025, 61(10): 154-163.

Changchun Zhang,Xingchang Yang,Guohua Wang,Bingjing Wang,Shijie Li,Fangzhou Chen,Yongtai Ge,Juan Shi,Junguo Zhang. A Detection Method of Alien Forest Pests with Integrating Multi-scale Attention Features[J]. Scientia Silvae Sinicae, 2025, 61(10): 154-163.

图/表 9

图1

图2

表1

表2

图3

图4

表3

表4

图5

参考文献 0

	陈万志, 袁　航. 改进YOLOv8n的林业害虫检测方法. 北京林业大学学报, 2025, 47 (2): 119- 131. doi: 10.12171/j.1000-1522.20240326
	Chen W Z, Yuan H. Forestry pest detection method based on improved YOLOv8n. Journal of Beijing Forestry University, 2025, 47 (2): 119- 131. doi: 10.12171/j.1000-1522.20240326
	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. 世界林业研究, 2024, 37 (4): 53- 57.
	Cui X C, Lei Y D. Research progress in deep-learning-based intelligent forest pest detection methods. World Forestry Research, 2024, 37 (4): 53- 57.
	宫　妍, 翟俊杰, 王　凯, 等. 基于改进YOLOv7的玉米作物害虫检测研究. 计算机测量与控制, 2024, 32 (9): 58- 65.
	Gong Y, Zhai J J, Wang K, et al. Study on corn crop pest detection based on improved YOLOv7. Computer Measurement & Control, 2024, 32 (9): 58- 65.
	郭嘉璇, 王蓉芳, 南江华, 等. 融入全局相应归一化注意力机制的YOLOv5农作物害虫识别模型. 农业工程学报, 2024, 40 (8): 159- 170. doi: 10.11975/j.issn.1002-6819.202310226
	Guo J X, Wang R F, Nan J H, et al. YOLOv5 model integrated with GRN attention mechanism for insect pest recognition. Transactions of the Chinese Society of Agricultural Engineering, 2024, 40 (8): 159- 170. doi: 10.11975/j.issn.1002-6819.202310226
	何　颖, 陈丁号, 彭　琳. 基于改进YOLOv5模型的经济林木虫害目标检测算法研究. 中国农机化学报, 2022, 43 (4): 106- 115.
	He Y, Chen D H, Peng L. Research on objeet detection algorithm of economic forestry pests based on improved YOLOv5. Journal of Chinese Agricultural Mechanization, 2022, 43 (4): 106- 115.
	匡敏球, 李　旭, 陈　熵, 等. 基于改进YOLOv8n的轻量化辣椒花目标检测方法. 农业工程学报, 2025, 41 (12): 198- 207. doi: 10.11975/j.issn.1002-6819.202502114
	Kuang M Q, Li X, Chen S, et al. Lightweight chili flower target detection method based on improved YOLOv8n. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2025, 41 (12): 198- 207. doi: 10.11975/j.issn.1002-6819.202502114
	孙海燕, 陈云博, 封丁惟, 等. 基于注意力模型和轻量化YOLOv4的林业害虫检测方法. 计算机应用, 2022, 42 (11): 3580- 3587. doi: 10.11772/j.issn.1001-9081.2021122164
	Sun H Y, Chen Y B, Feng D W, et al. Forest pest detection method based on attention model and lightweight YOLOv4. Journal of Computer Applications, 2022, 42 (11): 3580- 3587. doi: 10.11772/j.issn.1001-9081.2021122164
	孙丽萍, 谭少亨, 周宏威, 等. 基于YOLOv5的林业有害生物检测与识别. 森林工程, 2022, 38 (5): 104- 109. doi: 10.3969/j.issn.1006-8023.2022.05.013
	Sun L P, Tan S H, Zhou H W, et al. Forestry pests detection and identification based on YOLOv5. Forest Engineering, 2022, 38 (5): 104- 109. doi: 10.3969/j.issn.1006-8023.2022.05.013
	王俊岭, 王晨晨, 熊玉华. 基于YOLOv5l-Im的排水管道缺陷检测方法及效果分析. 科学技术与工程, 2024, 24 (18): 7833- 7842. doi: 10.12404/j.issn.1671-1815.2304688
	Wang J L, Wang C C, Xiong Y H. Detection and effectiveness of drainage pipeline defects based on YOLOv5l-Im. Science Technologr and Engineering, 2024, 24 (18): 7833- 7842. doi: 10.12404/j.issn.1671-1815.2304688
	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. 农业机械学报, 2024, 55 (11): 39- 48. doi: 10.6041/j.issn.1000-1298.2024.11.004
	Wang T H, Guo Y Z, Zhang J L, et al. Research on rice pest identification based on improved YOLOv5s. Transactions of the Chinese Society for Agricultural Machinery, 2024, 55 (11): 39- 48. doi: 10.6041/j.issn.1000-1298.2024.11.004
	王中天, 邹颖波, 吴昌霖, 等. 超越单一感知的农田害虫检测算法MRA-YOLOX. 计算机工程与应用, 2024, 60 (16): 206- 216. doi: 10.3778/j.issn.1002-8331.2305-0318
	Wang Z T, Zou Y B, Wu C L, et al. MRA-YOLOX for pest detection in farmland beyond single perception. Computer Engineering and Applications, 2024, 60 (16): 206- 216. doi: 10.3778/j.issn.1002-8331.2305-0318
	谢婉滢, 刘文萍, 王　晗. 无人机松林图像早期松材线虫病害检测. 林业科学, 2024, 60 (9): 124- 133. doi: 10.11707/j.1001-7488.LYKX20220898
	Xie W Y, Liu W P, Wang H. UAV images of pine forests for early detection of pine wood nematode infestation. Scientia Silvae Sinicae, 2024, 60 (9): 124- 133. doi: 10.11707/j.1001-7488.LYKX20220898
	杨红云, 肖小梅, 黄　琼, 等. 基于卷积神经网络和迁移学习的水稻害虫识别. 激光与光电子学进展, 2022, 59 (16): 333- 340.
	Yang H Y, Xiao X M, Huang Q, et al. Rice pest identification based on convolutional neural network and transfer learning. Progress in Laser and Optoelectronics, 2022, 59 (16): 333- 340.
	张长春, 李大方, 张军国. 基于Wasserstein 距离和相关对齐迁移学习的野生动物图像识别方法. 林业科学, 2024, 60 (8): 25- 32. doi: 10.11707/j.1001-7488.LYKX20230399
	Zhang C C, Li D F, Zhang J G. Wildlife images recognition method based on Wasserstein distance and correlation alignment transfer learning. Scient ia Silvae Sinicae, 2024, 60 (8): 25- 32. doi: 10.11707/j.1001-7488.LYKX20230399
	周志飞, 李　华, 冯毅雄, 等. 轻量化深度卷积神经网络设计研究进展. 计算机工程与应用, 2024, 60 (22): 1- 17. doi: 10.3778/j.issn.1002-8331.2404-0372
	Zhou Z F, Li H, Feng Y X, et al. Research progress on designing lightweight deep convolutional neural networks. Computer Engineering and Applications, 2024, 60 (22): 1- 17. doi: 10.3778/j.issn.1002-8331.2404-0372
	Chu J, Li Y, Feng H, et al. Research on multi-scale pest detection and identification method in granary based on improved YOLOv5. Agriculture, 2023, 13 (2): 364. doi: 10.3390/agriculture13020364
	He K, Zhang X, Ren S, et al. 2016. Deep residual learning for image recognition. IEEE Conference on Computer Vision and Pattern Recognition, 770−778.
	Hou Q, Zhou D, Feng J. 2021. Coordinate attention for efficient mobile network design. IEEE Conference on Computer Vision and Pattern Recognition, 13708−13717.
	Simonyan K, Zisserman A. 2015. Very deep convolutional networks for large-scale image recognition. International Conference on Learning Representations, 1234−1256.
	Wang J, Li Y, Feng H, et al. Common pests image recognition based on deep convolutional neural network. Computers and Electronics in Agriculture, 2020, 179, 105834. doi: 10.1016/j.compag.2020.105834
	Yulita I, Prabuwono A, Ardiansyah F, et al. Pest detection in agricultural farms using squeezeNet and multi-layer perceptron model. International Journal of Advanced Computer Science and Applications, 2024, 15 (6): 802- 808.
	Zhang Y, Ren W, Zhang Z, et al. 2022. Focal and efficient IOU loss for accurate bounding box regression. IEEE Conference Computer Vision and Pattern Recognition, 247−256.

模型 Models	精确率 Precision	召回率 Recall	平均精度均值 mAP@0.5	平均精度均值 mAP@.5:0.95
SSD	61.3	59.4	60.8	43.4
YOLOv4	71.5	76.3	75.2	62.5
YOLOv7	74.9	80.2	81.7	66.7
YOLOv8	76.4	81.2	82.5	67.3
YOLOv5s	74.3	79.1	79.2	64.9
MAF-YOLO	75.8	83.4	80.3	66.5

试验组 Experimental group	模型 Models	精确率 Precision	召回率 Recall	平均精度均值 mAP@0.5	平均精度均值 mAP@ 0.5:0.95
无域适应 Without domain adaptation	YOLOv5s	74.3	79.1	79.2	64.9
无域适应 Without domain adaptation	MAF-YOLO	75.8	83.4	82.8	67.5
有域适应 With domain adaptation	YOLOv5s	76.8	79.9	80.0	66.0
有域适应 With domain adaptation	MAF-YOLO	77.9	84.5	83.5	69.8

物种 Species	模型准确率 Model accuracy （%）
物种 Species	MAF-YOLO	YOLOv5s	YOLOv8	YOLOv7	YOLOv4
圆木蠹象Pissodes rotundatus	60.5	56.3	61.4	61.7	54.5
松六星吉Chrysobothris solieri	94.5	90.2	96.0	95.2	87.6
长林小蠹Hylurgus ligniperda	90.8	88.6	90.7	91.1	86.3
美国白蛾Hyphantria cunea	97.2	96.3	98.9	97.5	94.4
条带木蠹象Pissodes castaneus	40.6	34.5	38.6	37.9	34.8
黑腹尼虎天牛 Neoclytus acuminatus	87.4	84.4	87.0	86.3	83.3
桉天牛Phoracantha semipunctata	91.2	88.3	91.9	90.8	86.9
美松齿小蠹Ips pini	82.5	80.0	81.7	80.9	79.1
耐猛木蠹象Pissodes nemorensis	44.3	39.6	42.5	41.6	38.3
云杉木蠹象Pissodes piceae	42.7	37.9	41.3	39.8	37.5
松树木蠹象Pissodes pini	51.8	48.5	51.0	52.2	47.3
菲利普木蠹象Pissodes piniphilus	34.4	30.2	32.6	31.9	29.0
欧洲榆小蠹 Scolytus multistriatus	75.7	72.8	75.4	74.5	70.2
松树蜂Sirex noctilio	88.9	86.8	90.1	89.2	85.7
暗褐断眼天牛Tetropium fuscum	90.1	88.2	91.2	90.0	86.8

YOLOv5s基线网络 YOLOv5s baseline network	坐标注意力 Coordinate attention	小目标检测头 Small target detection head	Focal-EIoU损失函数 Focal-EIoU loss function	精确率 Precision	召回率 Recall	平均精度均值 mAP@0.5	平均精度均值 mAP@.5:.95
√	√	?	?	74.9	81.2	79.5	66.3
√	?	√	?	74.8	80.3	79.7	65.9
√	?	?	√	74.4	81.7	79.3	65.1
√	√	√	?	75.7	82.0	80.0	66.7
√	√	?	√	75.1	82.3	79.9	66.0
√	?	?	?	74.3	79.1	79.2	64.9
√	√	√	√	75.8	83.4	80.3	66.5

[1]	赵恩庭,张长春,赵海涛,张军国. 基于对抗学习的野生动物图像域适应识别方法[J]. 林业科学, 2025, 61(4): 1-8.
[2]	杨文翰,刘天宇,周俊池,胡文武,蒋蘋. 基于改进YOLOv5s的CNN-Swin Transformer森林野生动物图像目标检测算法[J]. 林业科学, 2024, 60(3): 121-130.

融合多尺度注意力特征的林业外来害虫检测方法

A Detection Method of Alien Forest Pests with Integrating Multi-scale Attention Features

RichHTML

PDF

可视化

摘要/Abstract

引用本文

使用本文

图/表 9

参考文献 0

相关文章 2

编辑推荐

Metrics

本文评价

作者中心

期刊获奖

引证指标

联系方式