基于对比学习的植物叶片病害识别

doi:10.3969/j.issn.1004-1524.20230093

浙江农业学报 ›› 2024, Vol. 36 ›› Issue (1): 215-224.DOI: 10.3969/j.issn.1004-1524.20230093

基于对比学习的植物叶片病害识别

杨新宇¹(), 冯全¹^,^*(), 张建华², 杨森¹

1.甘肃农业大学机电工程学院,甘肃兰州 730070
2.中国农业科学院农业信息研究所,北京 100081

收稿日期:2023-01-30 出版日期:2024-01-25 发布日期:2024-02-18
作者简介:杨新宇(1996—),女,河北秦皇岛人,硕士研究生,研究方向为图像处理。E-mail:yxy_0627@163.com
通讯作者: * 冯全,E-mail: fquan@sina.com
基金资助:
国家自然科学基金(32160421);国家自然科学基金(31971792);国家自然科学基金(32201663);甘肃省高等学校产业支撑计划(2021CYZC-57)

Plant leaf disease identification based on contrastive learning

YANG Xinyu¹(), FENG Quan¹^,^*(), ZHANG Jianhua², YANG Sen¹

1. Mechanical and Electrical Engineering College, Gansu Agricultural University, Lanzhou 730070, China
2. Agricultural Information Institute, Chinese Academy of Agricultural Sciences, Beijing 100081, China

Received:2023-01-30 Online:2024-01-25 Published:2024-02-18

摘要/Abstract

摘要：

目前,基于图像处理的植物病害识别多依赖于人工标注的卷积神经网络。本文基于自监督对比学习不依赖标签和大量数据即可实现独自学习的优势,研究了MoCo-v2、DeepCluster-v2、SwAV、BYOL这4种对比学习方法对植物叶片病害的识别效果,通过设置不同试验条件,使用PlantVillage开源数据集和自建的棉花病害数据集分别测试4种对比学习方法所训练的ResNet50编码器在Linear和Finetune两种模式下的病害识别效果,评估对比学习方法在植物叶片病害识别上的可行性。结果表明,在PlantVillage数据集中,Finetune模式的平均准确率略高于Linear,4种方法训练的编码器的平均识别准确率最高达99.83%,其中,DeepCluster-v2、BYOL在Finetune模式下的识别率最高,均为99.87%。在棉花病害数据集上,Finetune模式的效果略逊于Linear,DeepCluster-v2在Linear模式下获得最高识别准确率(98.86%)。整体来看,基于对比学习方法的病害识别率优于有监督模型的学习效果,在植物叶片病害识别领域展现出良好的应用前景。

关键词: 对比学习, 病害识别, 图像处理

Abstract:

At present, the recognition of plant disease via image processing mostly relies on the manually labeled convolutional neural network. However, the self-monitoring contrastive learning could achieve independent learning without relying on labels and large amounts of data. In view of this advantage, the effect of four contrastive learning methods, MoCo-v2, DeepCluster-v2, SwAV and BYOL, on the identification of plant leaf diseases were compared by setting different experimental conditions on the open-source dataset of PlantVillage and the self-built cotton disease dataset. The ResNet50 encoder trained by four contrastive learning methods was tested for disease identification both under Linear and Finetune modes, and the feasibility of the contrastive learning methods in identifying plant leaf diseases was evaluated. It was shown that the average accuracy under Finetune mode on the PlantVillage dataset was higher than that under Linear mode, and the highest identificaiton accuracy of the encoders trained by the four methods reached 99.83%. DeepCluster-v2 and BYOL had the highest identification rate under Finetune mode, both of which were 99.87%. On the self-built cotton disease dataset, the performance under Finetune mode was poorer than that under Linear mode, and the highest idenficaiton accuracy of DeepCluster-v2 under Linear mode was 98.86%. Overall, the disease identification rate based on contrastive learning method was superior to the supervised models, demonstrating good application prospects in plant leaf disease identification.

Key words: contrastive learning, disease identification, image processing

中图分类号:

TP391.41

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.

YANG Xinyu, FENG Quan, ZHANG Jianhua, YANG Sen. Plant leaf disease identification based on contrastive learning[J]. Acta Agriculturae Zhejiangensis, 2024, 36(1): 215-224.

图/表 8

参考文献 19

[1]	孙俊, 谭文军, 毛罕平, 等. 基于改进卷积神经网络的多种植物叶片病害识别[J]. 农业工程学报, 2017, 33(19): 209-215.
	SUN J, TAN W J, MAO H P, et al. Recognition of multiple plant leaf diseases based on improved convolutional neural network[J]. Transactions of the Chinese Society of Agricultural Engineering, 2017, 33(19): 209-215. (in Chinese with English abstract)
[2]	MOHANTY S P, HUGHES D P, SALATHÉ M. Using deep learning for image-based plant disease detection[J]. Frontiers in Plant Science, 2016, 7: 1419.
[3]	刘洋, 冯全, 王书志. 基于轻量级CNN的植物病害识别方法及移动端应用[J]. 农业工程学报, 2019, 35(17): 194-204.
	LIU Y, FENG Q, WANG S Z. Plant disease identification method based on lightweight CNN and mobile application[J]. Transactions of the Chinese Society of Agricultural Engineering, 2019, 35(17): 194-204. (in Chinese with English abstract)
[4]	侯志松, 冀金泉, 李国厚, 等. 集成学习与迁移学习的作物病害图像识别算法[J]. 中国科技论文, 2021, 16(7): 708-714.
	HOU Z S, JI J Q, LI G H, et al. Crop disease image recognition algorithm based on ensemble learning and transfer learning[J]. China Sciencepaper, 2021, 16(7): 708-714. (in Chinese with English abstract)
[5]	刘翱宇, 吴云志, 朱小宁, 等. 基于深度残差网络的玉米病害识别[J]. 江苏农业学报, 2021, 37(1): 67-74.
	LIU A Y, WU Y Z, ZHU X N, et al. Corn disease recognition based on deep residual network[J]. Jiangsu Journal of Agricultural Sciences, 2021, 37(1): 67-74. (in Chinese with English abstract)
[6]	郭小清, 范涛杰, 舒欣. 基于改进Multi-Scale AlexNet的番茄叶部病害图像识别[J]. 农业工程学报, 2019, 35(13): 162-169.
	GUO X Q, FAN T J, SHU X. Tomato leaf diseases recognition based on improved Multi-Scale AlexNet[J]. Transactions of the Chinese Society of Agricultural Engineering, 2019, 35(13): 162-169. (in Chinese with English abstract)
[7]	赵立新, 侯发东, 吕正超, 等. 基于迁移学习的棉花叶部病虫害图像识别[J]. 农业工程学报, 2020, 36(7): 184-191.
	ZHAO L X, HOU F D, LYU Z C, et al. Image recognition of cotton leaf diseases and pests based on transfer learning[J]. Transactions of the Chinese Society of Agricultural Engineering, 2020, 36(7): 184-191. (in Chinese with English abstract)
[8]	ALBELWI S. Survey on self-supervised learning: auxiliary pretext tasks and contrastive learning methods in imaging[J]. Entropy, 2022, 24(4): 551.
[9]	张重生, 陈杰, 李岐龙, 等. 深度对比学习综述[J]. 自动化学报, 2023, 49(1): 15-39.
	ZHANG C S, CHEN J, LI Q L, et al. Deep contrastive learning: a survey[J]. Acta Automatica Sinica, 2023, 49(1): 15-39. (in Chinese with English abstract)
[10]	CARON M, MISRA I, MAIRAL J, et al. Unsupervised learning of visual features by contrasting cluster assignments[C]// Proceedings of the 34th International Conference on Neural Information Processing Systems. December 6-12, 2020, Vancouver, BC, Canada. New York: ACM, 2020: 9912-9924.
[11]	GRILL J B, STRUB F, ALTCH’E F, et al. Bootstrap your own latent: a new approach to self-supervised learning[EB/OL].(2020-06-13) [2023-01-30]. https://arxiv.org/abs/2006.07733.
[12]	田浩江, 路娜, 崔二洋. 基于改进对比学习的道路裂缝图像分类[J]. 计算机系统应用, 2023, 32(2): 310-315.
	TIAN H J, LU N, CUI E Y. Road crack image classification based on improved contrastive learning[J]. Computer Systems and Applications, 2023, 32(2): 310-315. (in Chinese with English abstract)
[13]	奚琰. 基于对比学习的细粒度遮挡人脸表情识别[J]. 计算机系统应用, 2022, 31(11): 175-183.
	XI Y. Fine-grained occluded facial expression recognition based on contrastive learning[J]. Computer Systems and Applications, 2022, 31(11): 175-183. (in Chinese with English abstract)
[14]	GitHub Inc. linusericsson/ssl-transfer: code[EB/OL]. [2023-01-30]. https://github.com/linusericsson/ssl-transfer.
[15]	SOHN K. Improved deep metric learning with multi-class N-pair loss objective[EB/OL]. [2023-01-30]. https://papers.nips.cc/paper/2016/file/6b180037abbebea991d8b1232f8a8ca9-Paper.pdf.
[16]	陈伟文, 邝祝芳, 王忠伟. 基于卷积神经网络的种苗病害识别方法[J]. 中南林业科技大学学报, 2022, 42(7): 35-43.
	CHEN W W, KUANG Z F, WANG Z W. Method of seed disease recognition based on convolutional neural network[J]. Journal of Central South University of Forestry & Technology, 2022, 42(7): 35-43. (in Chinese with English abstract)
[17]	MOHAMETH F, CHEN B C, SADA K A. Plant disease detection with deep learning and feature extraction using plant village[J]. Journal of Computer and Communications, 2020, 8(6): 10-22.
[18]	TOO E C, LI Y J, NJUKI S, et al. A comparative study of fine-tuning deep learning models for plant disease identification[J]. Computers and Electronics in Agriculture, 2019, 161: 272-279.
[19]	ZHU D Q, FENG Q, ZHANG J H, et al. Cotton disease identification method based on pruning[J]. Frontiers in Plant Science, 2022, 13: 1038791.

模式 Mode	Beach_size	优化方法 Optimization method	各对比学习方法的准确率Accuracy of contrastive learning methods
模式 Mode	Beach_size	优化方法 Optimization method	MoCo-v2	DeepCluster-v2	SwAV	BYOL	平均值Mean
Linear	32	SGD+momentum	97.26	99.70	99.74	99.70	99.10
Linear	64	SGD+momentum	96.92	99.79	99.78	99.69	99.04
Finetune	32	SGD+momentum	99.13	99.75	99.71	99.66	99.56
Finetune	64	SGD+momentum	99.70	98.57	99.86	99.87	99.50
Finetune	32	SGD+momentum+Adam	99.52	99.77	99.64	99.59	99.63
Finetune	64	SGD+momentum+Adam	99.75	99.87	99.86	99.84	99.83

模式 Mode	Beach_size	优化方法 Optimization method	各对比学习方法的准确率Accuracy of contrastive learning methods
模式 Mode	Beach_size	优化方法 Optimization method	MoCo-v2	DeepCluster-v2	SwAV	BYOL	平均值Mean
Linear	32	SGD+momentum	97.26	99.70	99.74	99.70	99.10
Linear	64	SGD+momentum	96.92	99.79	99.78	99.69	99.04
Finetune	32	SGD+momentum	99.13	99.75	99.71	99.66	99.56
Finetune	64	SGD+momentum	99.70	98.57	99.86	99.87	99.50
Finetune	32	SGD+momentum+Adam	99.52	99.77	99.64	99.59	99.63
Finetune	64	SGD+momentum+Adam	99.75	99.87	99.86	99.84	99.83

类别 Category	准确率 Accuracy	类别 Category	准确率 Accuracy	类别 Category	准确率 Accuracy	类别 Category	准确率 Accuracy
C1	100.0	C11	99.54	C21	100.0	C31	99.73
C2	100.0	C12	100.0	C22	100.0	C32	99.50
C3	100.0	C13	100.0	C23	100.0	C33	99.12
C4	100.0	C14	100.0	C24	100.0	C34	99.68
C5	100.0	C15	100.0	C25	100.0	C35	99.68
C6	100.0	C16	100.0	C26	100.0	C36	99.91
C7	100.0	C17	100.0	C27	100.0	C37	100.0
C8	94.96	C18	100.0	C28	100.0	C38	100.0
C9	100.0	C19	100.0	C29	100.0
C10	96.28	C20	100.0	C30	98.98

类别 Category	准确率 Accuracy	类别 Category	准确率 Accuracy	类别 Category	准确率 Accuracy	类别 Category	准确率 Accuracy
C1	100.0	C11	99.54	C21	100.0	C31	99.73
C2	100.0	C12	100.0	C22	100.0	C32	99.50
C3	100.0	C13	100.0	C23	100.0	C33	99.12
C4	100.0	C14	100.0	C24	100.0	C34	99.68
C5	100.0	C15	100.0	C25	100.0	C35	99.68
C6	100.0	C16	100.0	C26	100.0	C36	99.91
C7	100.0	C17	100.0	C27	100.0	C37	100.0
C8	94.96	C18	100.0	C28	100.0	C38	100.0
C9	100.0	C19	100.0	C29	100.0
C10	96.28	C20	100.0	C30	98.98

模式 Mode	Beach_size	各对比学习方法的准确率Accuracy of contrastive learning methods
模式 Mode	Beach_size	MoCo-v2	DeepCluster-v2	SwAV	BYOL	平均值Mean
Linear	32	97.29	98.29	97.20	98.30	97.77
Linear	64	98.36	98.86	98.47	98.56	98.56
Finetune	32	96.38	96.56	96.38	96.23	96.39
Finetune	64	97.11	97.47	97.47	96.93	97.35

基于对比学习的植物叶片病害识别

Plant leaf disease identification based on contrastive learning

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 8

参考文献 19

相关文章 15

编辑推荐

Metrics

本文评价

类别 Category	精确率Precision				召回率Recall				F1值F1 score
类别 Category	M1	M2	M3	M4	M1	M2	M3	M4	M1	M2	M3	M4
D1	97.10	98.33	97.21	97.63	96.40	98.01	96.96	97.2	96.75	98.17	97.08	97.41
D2	97.85	97.17	97.81	97.44	99.21	98.10	99.01	98.7	98.53	97.63	98.41	98.07
D3	99.65	99.78	99.63	99.59	98.80	99.12	98.70	98.79	99.22	99.45	99.16	99.19
D4	99.59	99.60	99.67	99.64	100.00	100.00	100.00	100.00	99.79	99.80	99.83	99.82
D5	99.03	99.40	99.17	99.21	99.67	99.85	99.76	99.87	99.35	99.62	99.46	99.54
D6	87.78	89.66	88.00	87.79	86.22	86.67	85.56	85.19	86.99	88.14	86.76	86.47
D7	98.99	98.50	98.87	98.71	98.74	98.13	98.59	98.39	98.86	98.31	98.73	98.55
D8	99.83	99.67	99.79	99.75	99.65	100.00	99.79	99.75	99.74	99.83	99.79	99.75

[1]	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.
[2]	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.
[3]	张宁, 吴华瑞, 韩笑, 缪祎晟. 基于多尺度和注意力机制的番茄病害识别方法[J]. 浙江农业学报, 2021, 33(7): 1329-1338.
[4]	鲍烈, 王曼韬, 刘江川, 文波, 明月. 基于卷积神经网络的小麦产量预估方法[J]. 浙江农业学报, 2020, 32(12): 2244-2252.
[5]	刘志, 贺正, 苗芳芳, 贾彪. 基于无人机的水肥一体化玉米出苗率估算方法与试验[J]. 浙江农业学报, 2019, 31(6): 977-985.
[6]	步正延, 李臻峰, 宋飞虎, 李斌, 李静. 基于太赫兹成像技术的大豆叶片水分含量测定[J]. 浙江农业学报, 2018, 30(8): 1420-1426.
[7]	顾正敏, 李臻峰, 宋飞虎, 张君生, 庄为. 基于光场相机的大豆冠层叶面积无损测量方法研究[J]. 浙江农业学报, 2018, 30(12): 2144-2152.
[8]	张善文, 黄文准, 尤著宏. 基于物联网和深度卷积神经网络的冬枣病害识别方法[J]. 浙江农业学报, 2017, 29(11): 1868-1874.
[9]	李颀, 胡艺聪, 武付闯. 基于图像处理的育苗箱环境控制系统[J]. 浙江农业学报, 2017, 29(11): 1912-1919.
[10]	赵君爱1，2，魏艳春3. 基于改进中值滤波的图像噪声去除算法的研究[J]. 浙江农业学报, 2015, 27(6): 1078-.
[11]	石铁，索雪松*，周靖仁，韩昌兴，高立艾. 现代农业设施的溯源管理的研究 ——基于图像识别的太阳能电池片分析[J]. 浙江农业学报, 2015, 27(2): 285-.
[12]	李振1，2，廖同庆1，冯青春2，3，4，5，张东彦1，2，3，4，5，王秀2，3，4，5，*. 基于机器视觉的蔬菜种子活力指数检测算法研究及系统实现[J]. 浙江农业学报, 2015, 27(12): 2218-.
[13]	胡孟晗;董庆利;*;刘宝林;张重阳;叶飞. 基于椭圆傅里叶描述子的香蕉形状识别[J]. , 2014, 26(1): 200-205.
[14]	姚立健;边起;雷良育;赵大旭. 基于BP神经网络的水果分级研究[J]. , 2012, 24(5): 0-930.
[15]	刘建军;姚立健;彭樟林. 基于机器视觉的山核桃等级检测技术[J]. , 2010, 22(6): 854-858.