融合沙漏结构与改进坐标注意力的轻量级番茄叶片病害识别模型

doi:10.3969/j.issn.1004-1524.20240646

浙江农业学报 ›› 2025, Vol. 37 ›› Issue (1): 217-230.DOI: 10.3969/j.issn.1004-1524.20240646

融合沙漏结构与改进坐标注意力的轻量级番茄叶片病害识别模型

谷瑞¹^,²(), 宋翠玲¹, 钱春花³^,^*()

1.南京大学数字经济与管理学院,江苏南京 210008
2.苏州工业园区服务外包职业学院金融科技学院,江苏苏州 215123
3.苏州农业职业技术学院智慧农业学院,江苏苏州 215008

收稿日期:2024-07-21 出版日期:2025-01-25 发布日期:2025-02-14
作者简介:谷瑞(1982—),男,河南信阳人,硕士,副教授,研究方向为计算机视觉。E-mail:gur@siso.edu.cn
通讯作者: *钱春花,E-mail:chqian423@szai.edu.cn
基金资助:
2023年江苏省高职院校教师专业带头人高端研修项目(2023TDFX010);苏州市科技计划(SNG2023005);江苏现代农业产业技术体系项目(JATS-2023-348)

A lightweight tomato leaf disease recognition model integrating a sandglass structure with improved coordinate attention

GU Rui¹^,²(), SONG Cuiling¹, QIAN Chunhua³^,^*()

1. School of Digital Economy and Management, Nanjing University, Nanjing 210008, China
2. College of Financial Technology, Suzhou Industrial Park Institute of Services Outsourcing, Suzhou 215123, Jiangsu, China
3. College of Smart Agriculture, Suzhou Polytechnic Institute of Agriculture, Suzhou 215008, China

Received:2024-07-21 Online:2025-01-25 Published:2025-02-14

摘要/Abstract

摘要：

针对现有番茄叶片识别模型参数量大、计算复杂度高,推理时间长,难以部署在资源受限的移动设备上的问题,本文提出一种轻量级识别网络SG-ICA-MobileNetV3。首先,引入沙漏结构对MobileNetV3Small的倒残差块进行改造,在高维空间建立特征转换和跳跃连接缓解信息丢失问题,强化模型特征学习能力;其次,嵌入改进的坐标注意力机制,融合全局平均池化和最大池化自适应地学习不同位置的特征权重,增强对病害区域的感知能力;最后,将ReLU激活函数替换为ELU,缓解模型训练中梯度消失和权重偏置更新失效现象,提升网络收敛速度。结果表明,该模型在测试集上的分类准确率高达98.36%,在参数量、计算复杂度、推理速度、识别精度等方面优于MobileNetV3Small、MobileNeXt-1.0、MobileVit-S、ConvNeXt-V2等轻量级模型,并具有较强的泛化能力,能为快速、准确识别植物叶片病害提供算法支持。

关键词: 沙漏结构, SG-ICA-MobileNetV3, 坐标注意力, 番茄, 叶片, 病害识别

Abstract:

In response to the problems of large parameter sizes, high computational complexity, long inference time, and difficulty in deployment on resource-limited mobile devices associated with existing tomato leaf recognition models, this paper proposes a lightweight recognition network called SG-ICA-MobileNetV3. Firstly, a sandglass structure was introduced to modify the inverted residual blocks of MobileNetV3Small, establishing feature transformations and skip connections in high-dimensional space to mitigate information loss and strengthen the model’s feature learning capability. Secondly, an improved coordinate attention mechanism was embedded, integrating global average pooling and max pooling to adaptively learn feature weights at different positions, enhancing the perception ability of diseased areas. Finally, the ReLU activation function was replaced with ELU to alleviate gradient vanishing and the issue of weight bias updated failure during model training, improving network convergence speed. The results showed that the model achieved a classification accuracy of 98.36% on the test set. It outperformed lightweight models such as MobileNetV3Small, MobileNeXt-1.0, MobileVit-S, and ConvNeXt-V2 in terms of parameter count, computational complexity, inference speed, and recognition accuracy, demonstrating strong generalization capabilities. This study result could provide algorithm support for fast and accurate identification of plant leaf diseases.

Key words: sandglass structure, SG-ICA-MobileNetV3, coordinate attention, tomato, leaf, disease recognition

中图分类号:

谷瑞, 宋翠玲, 钱春花. 融合沙漏结构与改进坐标注意力的轻量级番茄叶片病害识别模型[J]. 浙江农业学报, 2025, 37(1): 217-230.

GU Rui, SONG Cuiling, QIAN Chunhua. A lightweight tomato leaf disease recognition model integrating a sandglass structure with improved coordinate attention[J]. Acta Agriculturae Zhejiangensis, 2025, 37(1): 217-230.

图/表 14

参考文献 26

[1]	马丽, 周巧黎, 赵丽亚, 等. 基于深度学习的番茄叶片病害分类识别研究[J]. 中国农机化学报, 2023, 44(7): 187-193.
	MA L, ZHOU Q L, ZHAO L Y, et al. Classification and recognition of tomato leaf diseases based on deep learning[J]. Journal of Chinese Agricultural Mechanization, 2023, 44(7): 187-193. (in Chinese with English abstract)
[2]	李大华, 仲婷, 王笋, 等. 基于改进ShuffleNet v2的轻量化番茄叶片病害识别[J]. 江苏农业科学, 2024, 52(3): 220-228.
	LI D H, ZHONG T, WANG Sun, et al. Lightweight disease identification for tomato leaves based on improved ShuffleNet v2[J]. Jiangsu Agricultural Sciences, 2024, 52(3): 220-228. (in Chinese with English abstract)
[3]	MOKHTAR U, ALI M A S, HASSENIAN A E, et al. Tomato leaves diseases detection approach based on Support Vector Machines[C]//2015 11th International Computer Engineering Conference (ICENCO), December 29-30, 2015, Cairo, Egypt. New York: IEEE, 2015: 246-250.
[4]	魏丽冉, 岳峻, 李振波, 等. 基于核函数支持向量机的植物叶部病害多分类检测方法[J]. 农业机械学报, 2017, 48 (S1):166-171.
	WEI L R, YUE J, LI Z B, et al. Multi-classification Detection Method of Plant Leaf Disease Based on Kernel Function SVM[J]. Transactions of the Chinese Society for Agricultural Machinery, 2017, 48 (S1):166-171.
[5]	XIE C Q, HE Y. Spectrum and image texture features analysis for early blight disease detection on eggplant leaves[J]. Sensors, 2016, 16(5): 676.
[6]	LIU B, ZHANG Y, HE D J. Identification of apple leaf diseases based on deep convolutional neural networks[J]. Symmetry-Base, 2018, 10(1):3390-3400.
[7]	DHANALAKSHMI R, BALAKRISHNAN K, SINHA B B. Tomato leaf disease identification by modified inception-based convolution neural networks[J]. The Imaging Science Journal, 2023, 71(5):408-424.
[8]	YU H L, CHENG X H, CHEN C C, et al. Apple leaf disease recognition method with improved residual network[J]. Multimedia Tools and Applications, 2022, 81(6): 7759-7782.
[9]	AHMED M, AHMED A. Palm tree disease detection and classification using residual network and transfer learning of inception-ResNet[J]. PLoS ONE, 2023, 18(3):1371-1384.
[10]	CHANG S L, YANG G J, CHENG J P. Recognition of wheat rusts in a field environment based on improved DenseNet[J]. Biosystems Engineering, 2023, 10(236): 10-21.
[11]	王杨, 李迎春, 许佳炜, 等. 基于改进Vision Transformer网络的农作物病害识别方法[J]. 小型微型计算机系统, 2024, 45(4): 887-893.
	WANG Y, LI Y C, XU J W, et al. Crop disease recognition method based on improved Vision Transformer network[J]. Journal of Chinese Computer Systems, 2024, 45(4): 887-893. (in Chinese with English abstract)
[12]	IANDOLA F N, HAN S, MOSKEWICZ M W, et al. SqueezeNet: AleNet level accuracy with 50 fewer parameters and<0.5 MB model size[EB/OL] (2016-06-01) [2022-11-11]. https://ariv.org/abs/1602.7360.
[13]	ZHANG X Y, ZHOU X Y, LIN M X, et al. ShuffleNet: an extremely efficient convolutional neural network for mobile devices[EB/OL]. (2017-12-07) [2024-06-01]. https://ar5iv.labs.arxiv.org/html/1707.01083.
[14]	TAN M X, LE Q V. EfficientNet: rethinking model scaling for convolutional neural networks[EB/OL]. (2019-09-11) [2024-06-01]. https://arxiv.org/abs/1905.11946.
[15]	HOWARD A G, ZHU M L, CHEN B, et al. MobileNets: efficient convolutional neural networks for mobile vision applications[EB/OL]. (2017-04-17) [2024-06-01]. http://arxiv.org/abs/1704.04861.
[16]	HAN K, WANG Y H, TIAN Q, et al. GhostNet:more features from cheap operations[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).June 13-19, 2020. Seattle, WA, USA. New York: IEEE, 2020: 1577-1586.
[17]	BI C K, WANG J M, DUAN Y L, et al. MobileNet based apple leaf diseases identification[J]. Mobile Networks and Applications, 2022, 27(1): 172-180.
[18]	HASSAM M, KHAN M A, ARMGHAN A, et al. A single stream modified MobileNet V3 and whale controlled entropy based optimization framework for citrus fruit diseases recognition[J]. IEEE Access, 2021, 10: 91828-91839.
[19]	袁培森, 欧阳柳江, 翟肇裕, 等. 基于MobileNetV3Small-ECA的水稻病害轻量级识别研究[J]. 农业机械学报, 2024, 55(1): 253-262.
	YUAN P S, OUYANG L J, ZHAI Z Y, et al. Lightweight identification of rice diseases based on improved ECA and MobileNetV3Small[J]. Transactions of the Chinese Society for Agricultural Machinery, 2024, 55(1): 253-262. (in Chinese with English abstract)
[20]	HU J, LI S, SUN G, et al. Squeeze-and-excitation networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020, 42(8): 2011-2023.
[21]	ZHOU D, HOU Q, CHEN Y, et al. Rethinking bottleneck structure for efficient mobile network design[C]// Computer Vision ECCV 2020: 16th European Conference, August 23-28, 2020, Glasgow, UK. Berlin:Springer-Verlag, 2020: 680-697.
[22]	HOU Q B, ZHOU D Q, FENG J S. Coordinate attention for efficient mobile network design[C]//2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). June 20-25, 2021. Nashville, TN, USA. IEEE, 2021: 13713-13722.
[23]	SANDLER M, HOWARD A, ZHU M L, et al. MobileNetV2: inverted residuals and linear bottlenecks[C]// 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. June 18-23, 2018. Salt Lake City, UT. IEEE, 2018: 4510-4520.
[24]	HOWARD A, SANDLER M, CHEN B, et al. Searching for MobileNetV3[C]//2019 IEEE/CVF International Conference on Computer Vision (ICCV). October 27-November 2, 2019. Seoul, Korea (South). IEEE, 2019: 1314-1324.
[25]	MEHTA S, RASTEGARI M. MobileViT: light-weight, general-purpose, and mobile-friendly vision transformer[EB/OL]. (2021-10-05) [2024-06-01]. https://arxiv.org/abs/2110.02178v2.
[26]	WANG S R, GAO J, LI Z M, et al. Closer look at self-supervised lightweight vision transformers[EB/OL]. (2023-05-03) [2024-06-01]. https://arxiv.org/abs/2205.14443.

输入 Input	操作 Operator	通道数 Exp_size	输出维度 Out	注意力 ICA	激活函数 Non-Linearity	步长 Stride
224×224×3	Conv2d,3×3	-	16	-	HS	2
112×112×16	SG-ICA bneck,3×3	16	16	√	ELU	2
56×56×16	SG-ICA bneck,3×3	72	24	-	ELU	2
28×28×24	SG-ICA bneck,3×3	88	24	-	ELU	1
28×28×24	SG-ICA bneck,5×5	96	40	√	HS	2
14×14×40	SG-ICA bneck,5×5	240	40	√	HS	1
14×14×40	SG-ICA bneck,5×5	240	40	√	HS	1
14×14×40	SG-ICA bneck,5×5	120	48	√	HS	1
14×14×48	SG-ICA bneck,5×5	144	48	√	HS	1
14×14×48	SG-ICA bneck,5×5	288	96	√	HS	2
7×7×96	SG-ICA bneck,5×5	576	96	√	HS	1
7×7×96	SG-ICA bneck,5×5	576	96	√	HS	1
7×7×96	Conv2d,1×1	-	576	√	HS	1
7×7×576	Pool,1×1	-	-	-	-	1
1×1×576	Conv2d,1×1,NBN	-	-	-	ELU	1
1×1×1024	Conv2d,1×1,NBN	-	-	-	HS	1

输入 Input	操作 Operator	通道数 Exp_size	输出维度 Out	注意力 ICA	激活函数 Non-Linearity	步长 Stride
224×224×3	Conv2d,3×3	-	16	-	HS	2
112×112×16	SG-ICA bneck,3×3	16	16	√	ELU	2
56×56×16	SG-ICA bneck,3×3	72	24	-	ELU	2
28×28×24	SG-ICA bneck,3×3	88	24	-	ELU	1
28×28×24	SG-ICA bneck,5×5	96	40	√	HS	2
14×14×40	SG-ICA bneck,5×5	240	40	√	HS	1
14×14×40	SG-ICA bneck,5×5	240	40	√	HS	1
14×14×40	SG-ICA bneck,5×5	120	48	√	HS	1
14×14×48	SG-ICA bneck,5×5	144	48	√	HS	1
14×14×48	SG-ICA bneck,5×5	288	96	√	HS	2
7×7×96	SG-ICA bneck,5×5	576	96	√	HS	1
7×7×96	SG-ICA bneck,5×5	576	96	√	HS	1
7×7×96	Conv2d,1×1	-	576	√	HS	1
7×7×576	Pool,1×1	-	-	-	-	1
1×1×576	Conv2d,1×1,NBN	-	-	-	ELU	1
1×1×1024	Conv2d,1×1,NBN	-	-	-	HS	1

方案 Schema	基线模型 Base	沙漏结构 SG	改进坐标注意力 ICA	激活函数 ELU	参数量 Params/MB	浮点运算数 FLOPs/10⁶	准确率 Accuracy/%
S0	√				2.40	68.2	93.25
S1	√	√			2.31	67.5	94.36
S2	√		√		2.19	66.6	94.42
S3	√			√	2.34	67.3	94.31
S4	√	√	√		2.28	65.1	96.58
S5	√		√	√	2.19	66.6	95.87
S6		√		√	2.31	67.3	96.53
S7	√	√	√	√	2.28	65.1	98.36

方案 Schema	基线模型 Base	沙漏结构 SG	改进坐标注意力 ICA	激活函数 ELU	参数量 Params/MB	浮点运算数 FLOPs/10⁶	准确率 Accuracy/%
S0	√				2.40	68.2	93.25
S1	√	√			2.31	67.5	94.36
S2	√		√		2.19	66.6	94.42
S3	√			√	2.34	67.3	94.31
S4	√	√	√		2.28	65.1	96.58
S5	√		√	√	2.19	66.6	95.87
S6		√		√	2.31	67.3	96.53
S7	√	√	√	√	2.28	65.1	98.36

类别 Category	准确率 Accuracy	精确率 Precision	召回率 Recall ratio	类别 Category	准确率 Accuracy	精确率 Precision	召回率 Recall ratio
健康叶片Healthy	100.00	100.00	100.00	红蜘蛛病Spider_mite	98.75	98.65	98.21
早疫病Early_blight	97.58	96.96	97.36	晚疫病Late_blight	96.55	96.98	96.06
花叶病Mosaic_virus	98.67	98.81	98.54	褐斑病Target_spot	96.95	96.14	96.26
叶霉病Leaf_mold	97.66	97.03	97.57	斑枯病Septoria_leaf_spot	97.54	97.76	97.11
黄花曲叶病Yellow_leaf_curl	98.66	98.04	97.56	细菌性斑点病Bacterial_spot	99.08	99.22	98.65

融合沙漏结构与改进坐标注意力的轻量级番茄叶片病害识别模型

A lightweight tomato leaf disease recognition model integrating a sandglass structure with improved coordinate attention

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 14

参考文献 26

相关文章 15

编辑推荐

Metrics

本文评价

模型 Model	参数量 Params/MB	浮点运算量 FLOPs/10⁶	模型大小 Model size/MB	准确率 Accuracy/%	召回率 Recall ratio/%
ShuffleNetV2^[13]	2.37	302.7	9.65	90.17	91.13
EfficientNet-B2^[14]	4.02	412.3	15.37	91.55	92.28
MobileNetV2-1.0^[23]	3.50	314.1	16.78	92.46	93.36
MobileNeXt-1.0^[20]	3.42	298.7	16.45	93.51	93.97
MobileNetV3Small^[24]	2.40	68.2	4.85	93.73	93.25
MobileVit-S^[25]	5.62	700.4	18.43	93.91	94.53
ConvNeXt-V2^[26]	5.21	380.6	21.26	94.29	94.84
SG-ICA-MobileNetV3	2.28	65.1	4.36	98.36	97.99

类别 Category	准确率 Accuracy/%	精确率 Precision/%	召回率 Recall ratio/%	平均推理时间 Average inference time/ms
斑点落叶病Alternaria_boltch	97.25	97.12	96.58	42.7
灰斑病Grey_spot	97.25	96.23	97.44	43.5
锈病Rust	96.32	96.37	96.26	44.6
褐斑病Brown_spot	98.14	98.27	98.45	41.4
白粉病Powdery_mildew	97.86	98.18	98.24	43.6
健康叶片Healthy	99.77	99.39	99.01	39.7
平均Average	97.76	97.59	97.66	42.58

[1]	李腾飞, 杨桂玲, 阮美颖, 褚田芬, 秦华, 邓美华. 不同肥药管理对设施番茄生产系统土壤健康与番茄性状的影响[J]. 浙江农业学报, 2025, 37(1): 145-158.
[2]	李大华, 孔舒, 李栋, 于晓. 基于改进SSD模型的柑橘叶片病害轻量化检测模型[J]. 浙江农业学报, 2024, 36(3): 662-670.
[3]	郭娜纳, 李伟, 黄立娟, 张涛, 魏兵强. 辣椒抗番茄斑萎病毒研究进展[J]. 浙江农业学报, 2024, 36(10): 2416-2425.
[4]	杨新宇, 冯全, 张建华, 杨森. 基于对比学习的植物叶片病害识别[J]. 浙江农业学报, 2024, 36(1): 215-224.
[5]	李必元, 岳智臣, 赵彦婷, 雷娟利, 胡齐赞, 陶鹏. 大白菜番茄红素β-环化酶基因BrLCYB的鉴定与功能分析[J]. 浙江农业学报, 2023, 35(9): 2090-2096.
[6]	刘媛媛, 王定坤, 邬雷, 黄德昌, 朱路. 基于知识蒸馏和模型剪枝的轻量化模型植物病害识别[J]. 浙江农业学报, 2023, 35(9): 2250-2264.
[7]	孙丽娟, 李世民, 郭焕仙, 金友帆, 李树萍, 董琼. 树番茄幼苗生长与氮磷钾化学计量特征对光照、肥料的响应[J]. 浙江农业学报, 2023, 35(8): 1793-1804.
[8]	张雪楠, 王乐乐, 钮铭轩, 詹妮, 任浩杰, 徐浩聪, 杨昆, 武立权, 柯健, 尤翠翠, 何海兵. 基于叶片反射光谱和叶绿素荧光估测水稻叶片含水量[J]. 浙江农业学报, 2023, 35(6): 1265-1277.
[9]	朱燕, 魏佳, 许自龙, 林天宝, 杨升, 刘岩, 吕志强, 刘培刚. 促生长植物激素对桑树叶片衰老过程生理生化指标的影响[J]. 浙江农业学报, 2023, 35(6): 1278-1285.
[10]	朱冬琴, 冯全, 张建华. 基于剪枝的植物病害识别方法[J]. 浙江农业学报, 2023, 35(6): 1462-1472.
[11]	娄茜棋, 梁燕. 五类不同果色番茄种质资源品质分析[J]. 浙江农业学报, 2023, 35(3): 582-589.
[12]	孙楠, 闫国超, 何勇, 朱祝军. 葫芦科作物叶片硅含量测定方法的优化[J]. 浙江农业学报, 2023, 35(2): 338-345.
[13]	李荣鹏, 买买提·沙吾提, 盛艳芳, 何旭刚. 基于CA-MobileNet-V2的核桃病害识别与应用[J]. 浙江农业学报, 2023, 35(12): 2977-2987.
[14]	杨迪, 张乃群, 王雪勇, 张军, 王新军. 基于数量性状的伏牛山野生中华猕猴桃资源综合评价[J]. 浙江农业学报, 2023, 35(10): 2354-2363.
[15]	魏茜雅, 梁腊梅, 林欣琪, 秦中维, 李映志. 褪黑素种子引发处理对干旱胁迫下朝天椒生长与生理特性的影响[J]. 浙江农业学报, 2023, 35(10): 2378-2388.