基于增强型轻量U-Net3+的茶叶病害诊断方法

doi:10.12133/j.smartag.SA202507010

Smart Agriculture ›› 2026, Vol. 8 ›› Issue (1): 15-27.doi: 10.12133/j.smartag.SA202507010

• 专题--农业病虫害智能识别与诊断 • 上一篇下一篇

基于增强型轻量U-Net3+的茶叶病害诊断方法

胡雨萌¹(), 关非凡¹, 谢东辰¹, 马萍¹, 余有本², 周杰², 聂炎明¹, 黄铝文¹^,³()

^1. 西北农林科技大学信息工程学院，陕西杨凌 712100，中国
^2. 西北农林科技大学园艺学院，陕西杨凌 712100，中国
^3. 陕西省农业信息智能感知与分析工程技术研究中心，陕西杨凌 712100，中国

收稿日期:2025-07-04 出版日期:2026-01-30
作者简介:
胡雨萌，硕士研究生，研究方向为生物图像处理。E-mail：yumenghu@nwafu.edu.cn
通信作者:
黄铝文，博士，副教授，研究方向为生物图像处理。E-mail：huanglvwen@nwafu.edu.cn

Tea Leaf Disease Diagnosis Based on Improved Lightweight U-Net3+

HU Yumeng¹(), GUAN Feifan¹, XIE Dongchen¹, MA Ping¹, YU Youben², ZHOU Jie², NIE Yanming¹, HUANG Lüwen¹^,³()

^1. College of Information Engineering, Northwest A&F University, Yangling 712100, China
^2. College of Horticulture, Northwest A&F University, Yangling 712100, China
^3. Key Laboratory of Agriculture Information Perception & Analytical Engineering and Technology Research Center, Yangling 712100, China

Received:2025-07-04 Online:2026-01-30
Foundation items:Science and Technology Project of the Ministry of Agriculture and Rural Affairs of China; 陕西省重点研发计划项目(2023-YBNY-219); 西北农林科技大学农业技术推广计划(Z222021411); 陕西省自然科学基础研究专项(2020JM-173)
About author:
HU Yumeng, E-mail: yumenghu@nwafu.edu.cn
Corresponding author:
HUANG Lüwen, E-mail: huanglvwen@nwafu.edu.cn

摘要/Abstract

摘要：

［目的/意义］ 茶叶病害常年影响着茶叶的产量和品质，针对既有茶叶病斑分割模型分割精细程度不足的问题，提出了一种茶叶病斑分割模型。 ［方法］ 提出了一种基于多尺度特征融合模块（Multi-scale Feature Fusion Module, MSFFM）、多尺度注意力机制（Dual Multi Scale Attention, DMSA）和条件随机场（Conditional Random Fields, CRF）的茶叶病斑分割模型MDC-U-Net3+。在U-Net3+的骨干网络中加入MSFFM获取病斑多个感受野下的特征信息，以减少编码器中特征的丢失；针对分割边界模糊问题，在跳跃连接过程中加入DMSA，充分融合全尺度下的细粒度和粗粒度语义信息；为进一步优化分割结果，利用CRF处理分割后的掩模图像。 ［结果和讨论］ 经验证，改进后模型平均像素精度（Mean Pixel Accuracy, mPA）为94.92%，平均交并比（Mean Intersection over Union, mIoU）为90.9%。相较于U-Net3+的mPA和mIoU分别提升了1.85和2.12个百分点，相对其他经典语义分割模型体现出了更优越的分割效果。 ［结论］ 本方法能够为病害自动检测与精准用药提供数据支持，减少病害造成的损失。

关键词: 病害诊断, 语义分割, U-Net3+, 多尺度特征融合, 注意力机制, 条件随机场

Abstract:

[Objective] Leaf diseases significantly affect both the yield and quality of tea throughout the year. To address the issue of inadequate segmentation finesse in the current tea spot segmentation models, a novel diagnosis of the severity of tea spots was proposed in this research, designated as MDC-U-Net3+, to enhance segmentation accuracy on the base framework of U-Net3+. [Methods] Multi-scale feature fusion module (MSFFM) was incorporated into the backbone network of U-Net3+ to obtain feature information across multiple receptive fields of diseased spots, thereby reducing the loss of features within the encoder. Dual multi-scale attention (DMSA) was incorporated into the skip connection process to mitigate the segmentation boundary ambiguity issue. This integration facilitates the comprehensive fusion of fine-grained and coarse-grained semantic information at full scale. Furthermore, the segmented mask image was subjected to conditional random fields (CRF) to enhance the optimization of the segmentation results [Results and Discussions] The improved model MDC-U-Net3+ achieved a mean pixel accuracy (mPA) of 94.92%, accompanied by a mean Intersection over Union (mIoU) ratio of 90.9%. When compared to the mPA and mIoU of U-Net3+, MDC-U-Net3+ model showed improvements of 1.85 and 2.12 percentage points, respectively. These results illustrated a more effective segmentation performance than that achieved by other classical semantic segmentation models. [Conclusions] The methodology presented herein could provide data support for automated disease detection and precise medication, consequently reducing the losses associated with tea diseases.

Key words: disease diagnosis, semantic segmentation, U-Net3+, multi-scale feature fusion, attention mechanism, conditional random fields

中图分类号:

TP183

胡雨萌, 关非凡, 谢东辰, 马萍, 余有本, 周杰, 聂炎明, 黄铝文. 基于增强型轻量U-Net3+的茶叶病害诊断方法[J]. 智慧农业(中英文), 2026, 8(1): 15-27.

HU Yumeng, GUAN Feifan, XIE Dongchen, MA Ping, YU Youben, ZHOU Jie, NIE Yanming, HUANG Lüwen. Tea Leaf Disease Diagnosis Based on Improved Lightweight U-Net3+[J]. Smart Agriculture, 2026, 8(1): 15-27.

图/表 16

参考文献 51

[1]	DING X L, LU Q, LI L P, et al. Measuring the impact of relative deprivation on tea farmers' pesticide application behavior: The case of Shaanxi, Sichuan, Zhejiang, and Anhui province, China[J]. Horticulturae, 2023, 9(3): 342.
[2]	HE M B, LI Y C, ZONG S X, et al. Life cycle assessment of carbon footprint of green tea produced by smallholder farmers in Shaanxi Province of China[J]. Agronomy, 2023, 13(2): 364.
[3]	PAYMODE A S, MALODE V B. Transfer learning for multi-crop leaf disease image classification using convolutional neural network VGG[J]. Artificial Intelligence in Agriculture, 2022, 6: 23-33.
[4]	YANG B, WANG Z L, GUO J Y, et al. Identifying plant disease and severity from leaves: A deep multitask learning framework using triple-branch Swin Transformer and deep supervision[J]. Computers and Electronics in Agriculture, 2023, 209: 107809.
[5]	SINGH A K, RAO A, CHATTOPADHYAY P, et al. Effective plant disease diagnosis using Vision Transformer trained with leafy-generative adversarial network-generated images[J]. Expert Systems with Applications, 2024, 254: 124387.
[6]	XU H, LI L F, YANG H, et al. Fine segmentation method for plant leaf disease spots based on deep learning[J]. Journal of Shenyang Agricultural University, 2025, 56(1): 92-107.
[7]	ZHOU Z M, ZHANG Y, GU Z H, et al. Deep learning approaches for object recognition in plant diseases: A review[J]. Intelligence & Robotics, 2023, 3(4): 514-537.
[8]	RITHARSON P I, RAIMOND K, MARY X A, et al. DeepRice: A deep learning and deep feature based classification of Rice leaf disease subtypes[J]. Artificial Intelligence in Agriculture, 2024, 11: 34-49.
[9]	HU G S, WAN M Z, WEI K, et al. Computer vision based method for severity estimation of tea leaf blight in natural scene images[J]. European Journal of Agronomy, 2023, 144: 126756.
[10]	NIE G, RAO H, LI Z, et al. Severity grading model for camellia oleifera anthracnose infection based on improved YOLACT[J]. Smart Agriculture, 2024, 6(3): 138-147.
[11]	NATH M, MITRA P, KUMAR D. A novel residual learning-based deep learning model integrated with attention mechanism and SVM for identifying tea plant diseases[J]. International Journal of Computers and Applications, 2023, 45(6): 471-484.
[12]	HUANG Z X, GOUDA M, YE S T, et al. Advanced deep learning algorithm for instant discriminating of tea leave stress symptoms by smartphone-based detection[J]. Plant Physiology and Biochemistry, 2024, 212: 108769.
[13]	SAJITHA P, ANDRUSHIA A D, ANAND N, et al. A review on machine learning and deep learning image-based plant disease classification for industrial farming systems[J]. Journal of Industrial Information Integration, 2024, 38: 100572.
[14]	ZHANG X X, LI F, ZHENG H Y, et al. UPFormer: U-sharped perception lightweight transformer for segmentation of field grape leaf diseases[J]. Expert Systems with Applications, 2024, 249: 123546.
[15]	ZHANG J Y, ZHANG B H, CHEN Z X, et al. A salient feature establishment tactic for cassava disease recognition[J]. Artificial Intelligence in Agriculture, 2024, 14: 115-132.
[16]	SHELHAMER E, LONG J, DARRELL T. Fully convolutional networks for semantic segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(4): 640-651.
[17]	LI X R, ZHOU H Q, CHANG Y Q, et al. A method of apple leaf spot segmentation based on improved DeepLabV3+[C]// International Conference on Cloud Computing, Performance Computing, and Deep Learning (CCPCDL 2023). Burlingame, California, USA: SPIE, 2023: 54.
[18]	BADRINARAYANAN V, KENDALL A, CIPOLLA R. SegNet: A deep convolutional encoder-decoder architecture for image segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(12): 2481-2495.
[19]	ZHANG S W, ZHANG C L. Modified U-Net for plant diseased leaf image segmentation[J]. Computers and Electronics in Agriculture, 2023, 204: 107511.
[20]	GONÇALVES J P, PINTO F A C, QUEIROZ D M, et al. Deep learning architectures for semantic segmentation and automatic estimation of severity of foliar symptoms caused by diseases or pests[J]. Biosystems Engineering, 2021, 210: 129-142.
[21]	LU J W, LU B B, MA W L, et al. EAIS-Former: An efficient and accurate image segmentation method for fruit leaf diseases[J]. Computers and Electronics in Agriculture, 2024, 218: 108739.
[22]	HU C, TAN L, WANG W, et al. Severity grading model for camellia oleifera anthracnose infection based on improved YOLACT[J]. Smart Agriculture, 2024, 6(5): 119-127.
[23]	ZHOU H, PENG Y T, ZHANG R P, et al. GS-DeepLabV3+: A mountain tea disease segmentation network based on improved shuffle attention and gated multidimensional feature extraction[J]. Crop Protection, 2024, 183: 106762.
[24]	CHEN T C, LI H X, CHEN J Z, et al. Detection network for multi-size and multi-target tea bud leaves in the field of view via improved YOLOv7[J]. Computers and Electronics in Agriculture, 2024, 218: 108700.
[25]	SUN Y G, WU F, GUO H P, et al. TeaDiseaseNet: Multi-scale self-attentive tea disease detection[J]. Frontiers in Plant Science, 2023, 14: 1257212.
[26]	HU G S, WANG H Y, ZHANG Y, et al. Detection and severity analysis of tea leaf blight based on deep learning[J]. Computers & Electrical Engineering, 2021, 90: 107023.
[27]	HU G S, WEI K, ZHANG Y, et al. Estimation of tea leaf blight severity in natural scene images[J]. Precision Agriculture, 2021, 22(4): 1239-1262.
[28]	SHWETHA V, BHAGWAT A, LAXMI V. LeafSpotNet: A deep learning framework for detecting leaf spot disease in jasmine plants[J]. Artificial Intelligence in Agriculture, 2024, 12: 1-18.
[29]	GUI Z Y, CHEN J N, LI Y, et al. A lightweight tea bud detection model based on YOLOv5[J]. Computers and Electronics in Agriculture, 2023, 205: 107636.
[30]	WU Y X, CHEN J N, WU S K, et al. An improved YOLOv7 network using RGB-D multi-modal feature fusion for tea shoots detection[J]. Computers and Electronics in Agriculture, 2024, 216: 108541.
[31]	MACDONALD W, SARI Y A, PAHLEVANI M. Grow-light smart monitoring system leveraging lightweight deep learning for plant disease classification[J]. Artificial Intelligence in Agriculture, 2024, 12: 44-56.
[32]	HUANG H M, LIN L F, TONG R F, et al. UNet 3+: A full-scale connected UNet for medical image segmentation[C]// ICASSP 2020 - 2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). Piscataway, New Jersey, USA: IEEE, 2020: 1055-1059.
[33]	AMRIT P, BARANWAL N, SINGH K N, et al. ConvNet-HIDE: Deep-learning-based dual watermarking for health-care images[J]. IEEE MultiMedia, 2024, 31(3): 78-87.
[34]	HAN Z M, JIAN M W, WANG G G. ConvUNeXt: An efficient convolution neural network for medical image segmentation[J]. Knowledge-Based Systems, 2022, 253: 109512.
[35]	ARDIMENTO P, AVERSANO L, BERNARDI M L, et al. Evo-GUNet3++: Using evolutionary algorithms to train UNet-based architectures for efficient 3D lung cancer detection[J]. Applied Soft Computing, 2023, 144: 110465.
[36]	GOKULKANNAN K, MOHANAPRAKASH T A, DAFNIROSE J, et al. Multiscale adaptive and attention-dilated convolutional neural network for efficient leukemia detection model with multiscale trans-res-Unet3+-based segmentation network[J]. Biomedical Signal Processing and Control, 2024, 90: 105847.
[37]	TIAN L L, ZHANG H X, LIU B, et al. VMF-SSD: A novel V-space based multi-scale feature fusion SSD for apple leaf disease detection[J]. IEEE/ACM Transactions on Computational Biology and Bioinformatics, 2023, 20(3): 2016-2028.
[38]	HUA Z W, QI L Z, DU D M, et al. Dual attention based multi-scale feature fusion network for indoor RGBD semantic segmentation[C]// 2022 26th International Conference on Pattern Recognition (ICPR). Piscataway, New Jersey, USA: IEEE, 2022: 3639-3644.
[39]	PENG P, WANG X Q, LI WP. Tea plant disease and pest forecasting and control techniques[M]. Beijing: China Agriculture Press, 2013.
[40]	RONNEBERGER O, FISCHER P, BROX T. U-Net: Convolutional networks for biomedical image segmentation[C]// Medical Image Computing and Computer-Assisted Intervention-MICCAI 2015. Cham, Germany: Springer, 2015: 234-241.
[41]	YI X M, ZHOU Y, WU P, et al. U-Net with coordinate attention and VGGNet: A grape image segmentation algorithm based on fusion pyramid pooling and the dual-attention mechanism[J]. Agronomy, 2024, 14(5): 925.
[42]	ZHOU Z W, RAHMAN SIDDIQUEE M M, TAJBAKHSH N, et al. UNet++: A nested U-Net architecture for medical image segmentation[C]// Deep Learning in Medical Image Analysis and Multimodal Learning for Clinical Decision Support. Cham, Germany: Springer, 2018: 3-11.
[43]	CHOUHAN S S, SINGH U P, JAIN S. Web facilitated anthracnose disease segmentation from the leaf of mango tree using radial basis function (RBF) neural network[J]. Wireless Personal Communications, 2020, 113(2): 1279-1296.
[44]	SAGAR A. DMSANet: Dual multi scale attention network[C]// Image Analysis and Processing-ICIAP 2022. Cham, Germany: Springer, 2022: 633-645.
[45]	HU G S, WU H Y, ZHANG Y, et al. A low shot learning method for tea leaf's disease identification[J]. Computers and Electronics in Agriculture, 2019, 163: 104852.
[46]	MAHESWARI P, RAJA P, HOANG V T. Intelligent yield estimation for tomato crop using SegNet with VGG19 architecture[J]. Scientific Reports, 2022, 12: 13601.
[47]	JI M M, WU Z B. Automatic detection and severity analysis of grape black measles disease based on deep learning and fuzzy logic[J]. Computers and Electronics in Agriculture, 2022, 193: 106718.
[48]	WANG X M, WU Z L, FANG C. TeaPoseNet: A deep neural network for tea leaf pose recognition[J]. Computers and Electronics in Agriculture, 2024, 225: 109278.
[49]	LI Y E, WANG Y, XU D Y, et al. An improved mask RCNN model for segmentation of 'kyoho' (Vitis labruscana) grape bunch and detection of its maturity level[J]. Agriculture, 2023, 13(4): 914.
[50]	LI K Y, ZHANG L X, LI B, et al. Attention-optimized DeepLab V3+ for automatic estimation of cucumber disease severity[J]. Plant Methods, 2022, 18(1): 109.
[51]	CHEN S, SONG Y Y, SU J Y, et al. Segmentation of field grape bunches via an improved pyramid scene parsing network[J]. International Journal of Agricultural and Biological Engineering, 2021, 14(6): 185-194.

Disease types	Pre-augmentation	Post-augmentation
Tea anthracnose （Gloeosporium theae-sinensis Miyake）	428	2 568
Tea leaf blight （Colletotrichum camelliae Massee）	375	2 250
Tea red scab （Cercospora theae BreadaDe Haan）	486	2 916
Tea blister blight （Exobasidium vexans Massee）	469	2 814
Tea red leaf spot （Phyllosticta theicola Petch）	435	2 610

Disease severity	Classification standard of leaf
Disease severity	Area of leaf spots	Number of leaf spots
0	none	0
1	less than 25% of the leaf area	1
2	25% to 50% of the leaf area	2
3	50% to 75% of the leaf area	3
4	more than 75% of the leaf area	4

Dataset division ratios	mPA/%	mIoU/%	Precision/%	Dice/%
8∶1∶1	93.73	89.20	94.63	93.78
7∶2∶1	94.20	90.26	95.28	94.29
7∶1∶2	94.90	90.88	95.24	94.56
6∶2∶2	94.92	90.90	95.24	94.58

Backbone network	mPA/%	mIoU/%
VGG16	93.35	88.84
VGG19	92.71	89.04
ResNet50	93.07	88.78
ResNet101	93.69	88.92
5-layer MSFFM	94.29	89.28

Model	mPA/%	mIoU/%	Precision/%	Dice/%	Frame rate/（f/s）	Average inference time/ms	Model size/MB
DeepLab V3+	88.59	84.38	94.67	89.85	22.476 2	44.898 6	10.987 4
PSPNet	91.91	84.81	86.71	88.78	9.601 5	44.499 6	9.601 5
U-Net	92.50	86.71	92.55	91.50	24.568 7	38.632 2	95.031 1
U-Net3+	93.07	88.78	94.59	92.96	20.633 6	48.576 1	143.031 8
MDC-U-Net3+	94.92	90.90	95.24	94.58	14.173 4	68.313 2	308.483 8

基于增强型轻量U-Net3+的茶叶病害诊断方法

Tea Leaf Disease Diagnosis Based on Improved Lightweight U-Net3+

在线阅读

知网下载

本地下载

可视化

摘要/Abstract

引用本文

使用本文

图/表 16

参考文献 51

相关文章 15

编辑推荐

Metrics

本文评价

Model	Frame rate/（frames/s）	Average training time/ms	Model size/MB
U-Net3+ （Baseline）	20.633 6	48.576 1	143.031 8
MSFFM+ U-Net3+	20.102 2	49.984 4	308.008 4
DMSA+MSFFM+ U-Net3+	16.813 7	57.489 6	308.483 8
MDC-U-Net3+	14.173 4	68.313 2	308.483 8

Original image
Label
Segmentation results of the proposed method
	Tea red leaf spot	Tea blister blight	Tea anthracnose	Tea leaf blight	Tea red scab

[1]	姚晓通, 曲绍业. 基于改进YOLOv12s的辣椒叶片病虫害轻量化检测方法[J]. 智慧农业(中英文), 2026, 8(1): 1-14.
[2]	蔡玉琴, 刘大铭, 徐琴, 李波洋, 刘博杰. 基于自适应Kalman滤波与GWO-LSTM-Attention的温室温湿度预测方法[J]. 智慧农业(中英文), 2026, 8(1): 148-155.
[3]	张运, 张露敏, 许广涛, 郝佳慧. 基于双分支与多尺度注意力机制的稻虾田遥感提取方法[J]. 智慧农业(中英文), 2025, 7(6): 185-195.
[4]	刘祎恒, 刘立波. 基于改进YOLOv12的遮挡环境下肉牛目标检测方法[J]. 智慧农业(中英文), 2025, 7(5): 182-192.
[5]	郭琪, 范艺璇, 闫新焕, 刘雪梅, 曹宁, 王震, 潘少香, 谭梦男, 郑晓冬, 宋烨. 多源数据融合技术在苹果无损检测中的应用研究进展[J]. 智慧农业(中英文), 2025, 7(4): 31-46.
[6]	王毅, 薛蓉, 韩文霆, 邵国敏, 侯艳巧, 崔茜彤. 基于CNN-LSTM-SA的玉米地上生物量估算[J]. 智慧农业(中英文), 2025, 7(4): 159-173.
[7]	彭秋珺, 李蔚然, 刘业强, 李振波. 基于改进HRNet的高精度鱼类姿态估计方法[J]. 智慧农业(中英文), 2025, 7(3): 160-172.
[8]	黎祖胜, 唐吉深, 匡迎春. 基于改进YOLOv10n的轻量化荔枝虫害小目标检测模型[J]. 智慧农业(中英文), 2025, 7(2): 146-159.
[9]	牛子昂, 裘正军. 基于改进YOLOv11-Pose的玉米植株骨架及表型参数提取方法[J]. 智慧农业(中英文), 2025, 7(2): 95-105.
[10]	吴六爱, 许雪珂. 基于改进YOLOv10n的轻量化番茄叶片病虫害检测方法[J]. 智慧农业(中英文), 2025, 7(1): 146-155.
[11]	权家璐, 陈雯柏, 王一群, 程佳璟, 刘亦隆. 基于GCN-BiGRU-STMHSA的农业干旱预测研究[J]. 智慧农业(中英文), 2025, 7(1): 156-164.
[12]	齐梓均, 牛当当, 吴华瑞, 张礼麟, 王仑峰, 张宏鸣. 基于双维信息与剪枝的中文猕猴桃文本命名实体识别方法[J]. 智慧农业(中英文), 2025, 7(1): 44-56.
[13]	芦碧波, 梁迪, 杨洁, 宋爱青, 皇甫尚卫. 基于改进ENet的复杂背景下山药叶片图像分割方法[J]. 智慧农业(中英文), 2024, 6(6): 109-120.
[14]	胡程喜, 谭立新, 王文胤, 宋敏. 基于改进DeepLabV3+的轻量化茶叶嫩芽采摘点识别模型[J]. 智慧农业(中英文), 2024, 6(5): 119-127.
[15]	年悦, 赵凯旋, 姬江涛. 基于改进DeepLabCut模型的奶牛滑蹄检测方法[J]. 智慧农业(中英文), 2024, 6(5): 153-163.