基于特征-边界协同建模的茶叶病虫害识别方法

A Tea Pest and Disease Identification Method Based on Feature-Boundary Collaborative Modeling

摘要

由于茶叶病虫害具有高相似性和小目标的特点，在复杂场景下的智能识别任务中极易出现漏检、误检以及定位不准确等问题，本文在YOLOv12网络框架基础上，提出了一种基于特征-边界协同建模的茶叶病虫害识别方法。首先，设计了并行分层密联残差混合机制，用以增强图像在不同感受野下的细节表达能力，进而提升模型对模糊边界和复杂纹理病斑的识别精度。然后，在特征-边界协同建模基础上，又设计了频域自适应调制策略，提升模型对高频病斑特征的感知响应能力。最后，构建了边界积分模糊Dice损失函数，有效增强模型对病斑目标轮廓的结构约束与几何拟合能力。实验结果表明，提出的改进模型在茶叶病虫害识别任务中表现出了优异的性能，识别精度、召回率和 mAP 分别达到了94.9%、98.6%和96.8%，相较原始YOLOv12分别提升了3.6%、3.5%和2.7%，且依然保持较高的实时性，推理速度可达45.7 f/s，为精准农业中的茶叶病虫害智能识别提供了可靠技术路径。

Abstract

Due to the high similarity and small target characteristics of tea pests and diseases, issues such as missed detection, false detection, and inaccurate positioning are prone to occur in intelligent recognition tasks under complex scenes. Based on the YOLOv12 network framework, this paper proposes a feature-boundary collaborative modeling method for tea pest and disease identification. Firstly, it designs a parallel hierarchical densely-connected residual mixing mechanism to enhance the detailed expression capacity of images in different receptive fields, thereby improving the model's recognition accuracy for blurred boundaries and complex-textured lesions. Then, based on the feature-boundary collaborative modeling, it designs a frequency-domain adaptive modulation strategy to enhance the model's perceptual response capacity to high-frequency lesion features. Finally, it constructs a boundary-integral fuzzy Dice loss function, effectively strengthening the model's structural constraints and geometric fitting capability for lesion target contours. The experimental results show that the proposed improved model exhibits outstanding performance in tea pest and disease identification tasks. The recognition precision, recall, and mAP reach 94.9%, 98.6%, and 96.8%, respectively, representing improvements of 3.6%, 3.5%, and 2.7% compared to the original YOLOv12. Moreover, the model maintains a high level of real-time performance, with an inference speed of 45.7 f/s. This provides a reliable technical pathway for intelligent identification of tea pests and diseases in precision agriculture.

关键词

茶叶病虫害识别 / YOLOv12 / 频域自适应调制 / 特征-边界协同建模 / 并行分层密联残差混合 / 边界积分模糊Dice损失

Key words

Tea pest and disease identification / YOLOv12 / frequency-domain adaptive modulation / feature-boundary collaborative modeling / parallel hierarchical densely-connected residual mixing / boundary-integral fuzzy dice loss

参考文献

[1] Bouacida I, Farou B, Djakhdjakha L, et al. Innovative deep learning approach for cross-crop plant disease detection: A generalized method for identifying unhealthy leaves[J]. Information Processing in Agriculture, 2025, 12(1): 54-67.

[2] Upadhyay A, Chandel N S, Singh K P, et al. Deep learning and computer vision in plant disease detection: a comprehensive review of techniques, models, and trends in precision agriculture[J]. Artificial Intelligence Review, 2025, 58(3): 1-64.

[3] Hussein Z J . Combining deep features and MSVM characteristics for enhanced classification of plant diseases[J]. Engineering, Technology and Applied Science Research, 2025, 15(4): 24442-24448.

[4] Morchid A, Marhoun M, El Alami R, et al. Intelligent detection for sustainable agriculture:a review of IoT-based embedded systems, cloud platforms, DL, and ML for plant disease detection[J]. Multimedia Tools and Applications, 2024, 83(28): 70961-71000.

[5] 乔世成, 党珊珊, 何海祝, 等. 基于卷积神经网络的农作物病害检测研究综述[J]. 山西农业大学学报(自然科学版), 2025, 45(02): 113-127.

[6] 方晓捷, 严李强, 张福豪, 等. 基于深度学习的农作物图像识别的发展[J]. 江苏农业科学, 2024, 52(20): 18-24.

[7] 曹英丽, 张弘泽, 郭福旭, 等. 基于无人机遥感的农作物病害监测研究进展[J]. 沈阳农业大学学报, 2024, 55(05): 616-628.

[8] Dolatabadian A, Neik T X, Danilevicz M F, et al. Image-based crop disease detection using machine learning [J]. Plant Pathology, 2025, 74(1): 18-38.

[9] Wang J, Chen Y, Huang J, et al. Leveraging machine learning for advancing insect pest control:A bibliometric analysis[J]. Journal of Applied Entomology, 2025, 149(3): 293-308.

[10] Abrol V, Gupta G, Pradhan S R . Smart farming: An integrated platform for crop and fertilizer prediction using machine learning and orange data mining[J]. SN Computer Science, 2025, 6(5): 1-16.

[11] Ahmad W, Azhar E, Anwar M, et al. Machine learning approaches for detecting vine diseases: A comparative analysis [J]. Journal of Informatics and Web Engineering, 2025, 4(1): 99-110.

[12] Aguirre-Rodríguez E Y, Rodriguez Gamboa A A, Aguirre Rodríguez E C, et al. Rice leaf disease detection using the stretched neighborhood effect color to grayscale method and machine learning[J]. Scientia Agropecuaria, 2025, 16(1): 123-136.

[13] Zhang Xiubin, Chen Hang, Huang Yongjian . Application research on detection of edge characteristics of tea garden disease based on UAV inspection [C]. Proc of the 2024 IEEE 2nd International Conference on Sensors, Electronics and Computer Engineering (ICSECE). Piscataway, NJ: IEEE Press, 2024: 248-253.

[14] 梁莹, 马琨, 张馨予, 等. 基于近红外光谱和随机森林的烟叶病害种类识别[J]. 激光与光电子学进展, 2024, 61(15): 362-369.

[15] Bhola A, Kumar P . Deep feature-support vector machine based hybrid model for multi-crop leaf disease identification in corn, rice, and wheat[J]. Multimedia Tools and Applications, 2025, 84(8): 4751-4771.

[16] 商炎亮, 吴凯, 周文. 基于多模态分层注意力网络的玉米病害叶片识别[J]. 中国农机化学报, 2025, 46(06): 162-168.

[17] Safari M M, Malian A . Plant disease mapping in paddy growing stages using remotely sensed data[J]. Environmental Earth Sciences, 2025, 84(1): 1-19.

[18] 刘宇平, 刘程飞, 赵平伟. 基于DeepLabv3-Faster R-CNN 的水稻叶片病害检测方法[J]. 中国农机化学报, 2025, 46(04): 108-113+132.

[19] Vo H T, Mui K C, Thien N N . Fusing xception and ResNet50 features for robust grape leaf disease classification [J]. Bulletin of Electrical Engineering and Informatics, 2025, 14(1): 268-275.

[20] 夏顺兴, 倪铭, 罗友璐, 等. 基于改进YOLOv8n的草莓叶片病害检测方法[J]. 江苏农业学报, 2025, 41(04): 664-675.

[21] 黄政, 张涛, 孔万仔, 等. 基于YOLOv8n 的梨树叶片病害检测模型[J]. 湖南农业大学学报(自然科学版), 2025, 51(02): 113-121.

[22] 郭文娟, 冯全. 基于改进YOLO的棉花叶片病害检测[J]. 干旱地区农业研究, 2024, 42(06): 195-205.

[23] Liu C T, Cai J H, Wu Z X, et al. Tea pest and disease detection algorithm based on deep learning [C]. Proc of the 2025 5th International Conference on Artificial Intelligence and Industrial Technology Applications (AIITA). Piscataway, NJ: IEEE Press, 2025: 2066-2070.

[24] Kwon H J, Suh S . Effective automatic weed detection with improved YOLOv10 [J]. Journal of The Korea Society of Computer and Information, 2024, 29(11): 89-96.

[25] Wang C, Han Y, Yang C, et al. CF-YOLO for small target detection in drone imagery based on YOLOv11 algorithm [J]. Scientific Reports, 2025, 15(1): 1-18.

基金资助

河南省高等学校重点科研项目计划(24B520025)

开封市科技发展计划项目(2402002)