AGLU-YOLO：轻量化柑橘叶片病害实时检测算法研究

doi:10.19788/j.issn.2096-6369.000145

摘要/Abstract

摘要：

针对柑橘叶部病害检测中精度不足、病斑细小易混淆、果园背景复杂等问题，提出一种轻量化检测算法（AGLU-YOLO）。该方法在主干网络C3k2模块中融合AdditiveBlock与卷积门控线性单元（CGLU）共同构成 C3k2_AdditiveBlock_CGLU 模块，前者以加性建模增强长程依赖与全局上下文表征，后者通过深度可分离3×3卷积与点卷积实现条件门控，抑制复杂纹理导致的误激活并强化小尺度病斑响应；同时在特征融合阶段加入AFCA注意力机制以提升跨层语义交互与多尺度鲁棒性。其次为满足边缘部署需求，采用LAMP分层重要度剪枝算法对通道/层级进行联合压缩，并进行轻量微调以恢复精度；随后将模型导出为ONNX并通过TensorRT实施算子融合与低精度推理优化，实现低时延、高吞吐的实时检测。通过在自制数据集实验验证，AGLU-YOLO在Precision、Recall与mAP@0.5指标上相较基线YOLOv11n分别提高3.1、4.4与5.1个百分点，且兼顾体积与识别速度，展现出更强的鲁棒性与多尺度病斑适应性，可有效满足柑橘叶病快速、准确识别的应用需求。

关键词: 柑橘叶病检测, 小目标检测, YOLOv11算法, 深度学习, CGLU

Abstract:

A lightweight detection algorithm (AGLU-YOLO) is proposed to solve the problems of insufficient accuracy, small lesions and complex orchard background in citrus leaf disease detection. This method fuses AdditiveBlock and Convolutional Gated Linear Unit (CGLU) in the C3k2 module of the backbone network to form the C3k2_AdditiveBlock_CGLU module. The former enhances long-range dependence and global context representation by additive modeling, and the latter realizes conditional gating by depthwise separable 3 × 3 convolution and point convolution to suppress false activation caused by complex texture and enhance small-scale lesion response. At the same time, the AFCA attention mechanism is added in the feature fusion stage to improve cross-layer semantic interaction and multi-scale robustness. Secondly, in order to meet the needs of edge deployment, the LAMP hierarchical importance pruning algorithm is used to jointly compress the channel / level, and a lightweight fine-tuning is performed to restore the accuracy; then the model is exported to ONNX and operator fusion and low-precision inference optimization are implemented through TensorRT to achieve real-time detection with low latency and high throughput. Through experimental verification on the self-made dataset, AGLU-YOLO improves the Precision, Recall and mAP @0.5 indexes by 3.1, 4.4 and 5.1 percentage points, respectively, compared with the baseline YOLOv11n, and takes into account the volume and recognition speed, showing stronger robustness and multi-scale lesion adaptability, which can effectively meet the application requirements of rapid and accurate identification of citrus leaf disease.

Key words: citrus leaf disease detection, small target detection, YOLOv11 algorithm, deep learning, CGLU

肖吟枫, 杨抒. AGLU-YOLO：轻量化柑橘叶片病害实时检测算法研究[J]. 农业大数据学报, 2026, 8(2): 141-154.

XIAO YinFeng, YANG Shu. AGLU-YOLO: Research on Real-time Detection Algorithm of Lightweight Citrus Leaf Disease[J]. Journal of Agricultural Big Data, 2026, 8(2): 141-154.

图/表 13

图1

图2

图3

图4

图5

图6

表1

表2

表3

表4

表5

表6

图7

参考文献 32

[1]	张勇. 基于深度学习的柑橘病虫害动态识别系统[D]. 成都: 西南民族大学, 2023.
	ZHANG Y. Dynamic recognition system for citrus pests and diseases based on deep learning[D]. Chengdu: Southwest Minzu University, 2023.
[2]	王海光. 植物病害图像识别及其发展前景. 植物病理学报, 2025, 55(4):964-977. doi: 10.13926/j.cnki.apps.000981
	WANG H G. Image recognition of plant diseases and its development prospects. Acta Phytopathologica Sinica, 2025, 55(4): 964-977.
[3]	唐琪. 基于多源数据的南方柑橘种植区适宜性评价与精准识别研究[D]. 赣州: 江西理工大学, 2025.
	TANG Q. Research on the suitability evaluation and precise identification of Southern Citrus Planting Areas based on multi-source data[D]. Ganzhou: Jiangxi University of Technology, 2025.
[4]	郑宇达, 陈仁凡, 杨长才, 等. 基于改进YOLOv5s模型的柑橘病虫害识别方法. 华中农业大学学报, 2024, 43(2):134‐143.
	ZHENG Y D, CHEN R F, YANG C C, et al. A method for citrus pest and disease recognition based on the improved YOLOv5s model. Journal of Huazhong Agricultural University, 2024, 43(2): 134-143.
[5]	曹静. 基于视频图像的大棚白菜叶部病害识别与预警系统设计与实现[D]. 保定: 河北农业大学, 2019.
	CAO J. Design and implementation of a system for identifying and warning of leaf diseases of green cabbage in greenhouses based on Video images[D]. Baoding: Hebei Agricultural University, 2019.
[6]	黄虹, 刘俊. 基于改进Faster RCNN的柑橘病虫害检测算法研究. 现代化农业, 2025(1):53-55.
	HUANG H, LIU J. Research on citrus pest and disease detection algorithm based on improved Faster RCNN. Modern Agriculture, 2025 (1): 53-55.
[7]	宋中山, 汪进, 郑禄, 等. 基于二值化的Faster R-CNN柑橘病虫害识别研究. 中国农机化学报, 2022, 43(6):150-158.
	SONG Z S, WANG J, ZHENG L, et al. Research on citrus pest and disease recognition based on binaryized Faster R-CNN. Chinese Journal of Agricultural Machinery Chemistry, 2022, 43(6): 150-158.
[8]	罗均. 基于深度学习的柑橘病害识别研究[D]. 武汉: 中南民族大学, 2021.
	LUO J. Research on citrus disease recognition based on deep learning[D]. Wuhan: South-Central Minzu University, 2021.
[9]	金雨纯, 甄元元, 刘平. 基于迁移改进MobileNetV3的玉米叶片病害识别研究. 中国农学通报, 2025, 41(23):145-154. doi: 10.11924/j.issn.1000-6850.casb2024-0773
	JIN Y C, ZHEN Y Y, LIU P. Research on corn leaf disease recognition based on improved MobileNetV3 with migration. Chinese Agricultural Science Bulletin, 2025, 41(23): 145-154.
[10]	吴世琛, 毛玉明, 胡慧中, 等. 基于改进YOLO11n的葡萄果叶病害检测方法. 农业工程学报, 2025, 41(14):140-147.
	WU S C, MAO Y M, HU H Z, et al. Detection method for grape leaf and fruit diseases based on improved YOLO11n. Transactions of the Chinese Society of Agricultural Engineering, 2025, 41(14): 140-147.
[11]	LI D, AHMED F, WU N, et al. Yolo-JD: A Deep Learning Network for jute diseases and pests detection from images. Plants, 2022, 11(7): 937. doi: 10.3390/plants11070937
[12]	CHEN Z, WU R, LIN Y, et al. Plant disease recognition model based on improved YOLOv5. Agronomy, 2022, 12: 365. doi: 10.3390/agronomy12020365
[13]	鲁子翱, 张婧婧, 韩博, 等. 基于改进YOLOv7-tiny的小麦麦穗检测方法. 江苏农业学, 2024.52(20):147-156.
	LU Z A, ZHANG J J, HAN B, et al. Detection method for wheat ear based on improved YOLOv7-tiny. Jiangsu Agricultural Sciences, 2024. 52(20): 147-156.
[14]	徐玮, 周佳良, 钱啸, 等. 基于AR眼镜和改进YOLOv8m-seg的田间小麦赤霉病严重度识别方法. 农业大数据学报, 2024, 6(4):497-508. doi: 10.19788/j.issn.2096-6369.000065
	XU W, ZHOU J L, QIAN X, et al. Method for identifying the severity of wheat fusarium wilt in the field based on AR glasses and improved YOLOv8m-seg. Journal of Agricultural Big Data, 2024, 6(4): 497-508.
[15]	吴丹, 马晓君, 刘德胜, 等. 基于改进YOLOv8的番茄目标检测算法研究. 农业大数据学报, 2025, 7(3):281-293. doi: 10.19788/j.issn.2096-6369.000075
	WU D, MA X J, LIU D S, et al. Research on tomato target detection algorithm based on improved YOLOv8. Journal of Agricultural Big Data, 2025, 7(3): 281-293.
[16]	ZHANG X, XUN Y, CHEN Y. Automated identification of citrus diseases in orchards using deep learning. Biosystems Engineering, 2022, 223: 249-258. doi: 10.1016/j.biosystemseng.2022.09.006
[17]	KANG J, ZHAO L J. Wheat pest detection based on PSA-YOLO11n. Journal of Agricultural Big Data, 2025, 7(3): 294-306. doi: 10.19788/j.issn.2096-6369.000101
[18]	ISMAIL R N, Armaneesa Naaman Hasoon, Israa Rafaa Abdulqader, et al. ROI - Enhancing detection of citrus disease based on YOLOv10. Mesopotamian Journal of Computer Science, 2025, 212: 316-328.
[19]	赵晓, 杨梦婷, 张懿丹. 基于改进YOLO v10n的苹果叶片病害检测方法. 农业机械学报, 2025, 56(9):657-666+728.
	ZHAO X, YANG M T, ZHANG Y D. Detection method for apple leaf diseases based on improved YOLO v10n. Transactions of the Chinese Society for Agricultural Machinery, 2025, 56(9): 657-666 + 728.
[20]	HE L, ZHOU Y, LIU L, et al. Application of the YOLOv11-seg algorithm for AI-based landslide detection and recognition. Scientific Reports, 2025, 15(1):12421-12421. doi: 10.1038/s41598-025-95959-y
[21]	冉书伟, 张丽丽, 王维政, 等. 基于改进YOLOv11的农作物实时检测研究. 智慧农业导刊, 2025, 5(10):27-31.
	RAN S W, ZHANG L L, WANG W Z, et al. Research on Real-time crop detection based on improved YOLOv11. Smart Agriculture Journal, 2025, 5(10): 27-31
[22]	PENG G, WANG K, MA J, et al. AGRI-YOLO: A lightweight model for corn weed detection with enhanced YOLOv11n. Agriculture, 2025, 15(18):1971. doi: 10.3390/agriculture15181971
[23]	KURNIAWAN W, KURNIASIH A, GHANI M A. Real or deepfake face detection in images and video data using YOLO11 algorithm. Journal of Artificial Intelligence and Engineering Application, 2025, 4(2): 1514-1521.
[24]	REN Z, LIU Y, YU T, et al. Simulated centrifugal fan blade fault diagnosis based on modulational depthwise Convolution-One- Dimensional Convolution Neural Network (MDC-1DCNN) model. Machines, 2025, 13(5): 356. doi: 10.3390/machines13050356
[25]	李彬, 李生林. 改进YOLOv11n的无人机小目标检测算法. 计算机工程与应用, 2025, 61(7):96-104. doi: 10.3778/j.issn.1002-8331.2411-0072
	LI B, LI S L. An improved algorithm for small object detection of unmanned aerial vehicles based on YOLOv11n. Computer Engineering and Applications, 2025, 61(7): 96-104.
[26]	HUANG J, WANG K, HOU Y, et al. LW-YOLO11: A lightweight arbitrary-oriented ship detection method based on improved YOLOv11. Sensors, 2025, 25: 65. doi: 10.3390/s25010065
[27]	CHEN D, XU K, SUN W, et al. REU-YOLO: A Context-Aware UAV-Based rice ear detection model for complex field scenes. Agronomy, 2025, 15(9): 2225. doi: 10.3390/agronomy15092225
[28]	ZHANG Y, GAO R, JI M, et al. TASNet-YOLO: An identification and classification model for surface defects of rough planed bamboo strips. Forests, 2025, 16(10):1595. doi: 10.3390/f16101595
[29]	DANG Y, WANG H, FAN Y. WAC-YOLO: A real-time traffic sign detection model for complex weather with feature degradation suppression and occlusion awareness. Journal of Real-Time Image Processing, 2025, 22(4): 155. doi: 10.1007/s11554-025-01739-2
[30]	GUO G, ZHANG L, JIA B, et al. Tool-YOLO: A target detection network based on feature extraction and feature fusion. Multimedia Systems, 2025, 31(6): 424. doi: 10.1007/s00530-025-02013-y
[31]	WANG S, FENG Y, JIN W, et al. DSEE-YOLO: A dynamic edge-enhanced lightweight model for infrared ship detection in complex maritime environments. Remote Sensing. 2025, 17(19):3325. doi: 10.3390/rs17193325
[32]	VINOTH K, SASIKUMAR P. Lightweight object detection in low light: pixel-wise depth refinement and TensorRT optimization. Results in Engineering, 2024, 23: 102510. doi: 10.1016/j.rineng.2024.102510

配置名称	参数
操作系统	Windows11
CPU	i7-12700H@2.3GHZ
GPU	NVIDIA GeForce RTX3070Ti
开发环境	Pycharm
语言	Python3.12
CUDA	12.9
优化器	Adam

基线模型	AddBlock	CGLU	AFCA	P/%	R/%	mAP@0.5/%	ModeSize/M	Parameters/M	FLOPs/G
YOLOv11n				92.1	85.3	90.2	5.3	2.5	6.3
	√			93.9	87.2	93.0	5.3	2.6	6.6
		√		93.8	86.4	92.8	4.6	2.2	5.6
			√	93.2	86.5	91.8	5.4	2.6	6.4
	√	√		94.8	88.9	94.7	5.3	2.4	6.1
	√	√	√	95.2	89.7	95.3	5.4	2.5	6.3

模型	P/%	R/%	mAP@0.5/%	Model Size/M	Params/M	FLOPs/G	FPS(帧/s)
YOLOv11n	92.1	85.3	90.2	5.30	2.5	6.3	200
AGLU-YOLO	95.2	89.7	95.3	5.40	2.5	6.3	195
Lamp剪枝后AGLU-YOLO	93.2	88.2	93.1	0.81	2.1	3.0	217

剪枝率	Params/M	FLOPs/G	Model Size/M	FPS	mAP@0.5/%
0	5.30	6.10	2.40	195	94.7
0.3	3.71	4.55	1.61	206	93.4
0.4	3.18	4.03	1.34	210	93.2
0.5	2.65	3.52	1.08	213	93.1
0.6	2.10	3.00	0.81	217	93.1
0.7	1.59	2.48	0.55	221	92.2

模型	mAP@0.5/%	单幅图像处理时间/ms	FPS(帧/s)
AGLU-YOLO	95.3	5.13	195
AGLU-YOLO+Lamp+TensorRT	93.8	3.53	283