基于迁移学习的玉米病害图像识别

doi:10.13304/j.nykjdb.2022.0218

摘要/Abstract

摘要：

传统的农作物病害检测多依靠人力和经验，信息化水平低。近年来，基于迁移学习的图像识别发展迅速，并在诸多领域取得了很好的应用效果。利用迁移学习的方法将MoblieNetV2模型对玉米病害图像数据集进行重新训练和微调，将优化后的玉米病害识别模型应用到移动端设备进行应用程序开发。结果表明，通过对预训练模型多次训练和微调，最终测试精度达到96.83%；最后利用优化后的模型开发了玉米病害识别应用APP，通过移动端APP对玉米进行拍照，进而获得诊断结果。该程序简单易操作，可以方便快速地识别玉米病害，在未来农业领域具有重要的应用价值。

关键词: 迁移学习, MobileNetV2, 图像识别, TensorFlow框架, 玉米病害

Abstract:

The traditional detection of crop disease mainly relies on manpower and experience， and the informatization level is low. In recent years， image recognition based on transfer learning has developed rapidly and achieved good application effect in many fields. MoblieNetV2 model was used to re-train and fine-tune corn disease image data set by transfer learning method. Then， the optimized corn disease recognition model was applied to the mobile terminal device for application development. The results showed that the final test accuracy reached to 96.83% after repeated training and fine-tuning of the pre-training model. Finally， the optimized model was used to develop a corn disease recognition APP， and the corn was photographed through the mobile APP to obtain the diagnosis results. The application was simple and easy to operate， which could facilitate and quickly identify maize diseases and have important application value in the future agricultural field.

Key words: transfer learning, MobileNetV2, image recognition, tensorflow framework, corn disease

中图分类号:

S126

张彦通, 苏前敏. 基于迁移学习的玉米病害图像识别[J]. 中国农业科技导报, 2023, 25(10): 119-125.

Yantong ZHANG, Qianmin SU. Image Recognition of Corn Disease Based on Transfer Learning[J]. Journal of Agricultural Science and Technology, 2023, 25(10): 119-125.

图/表 11

参考文献 15

1	龚瑞昆,刘佳.图像处理的玉米病害识别研究[J].现代电子技术,2021,44(24):149-152.
	GONG R K, LIU J. Research on recognition of corn diseases by image processing [J]. Mod. Electron. Technol., 2021, 44(24): 149-152.
2	刁智华,袁万宾,刁春迎,等.病害特征在作物病害识别中的应用研究综述[J].江苏农业科学,2019,47(5):71- 74.
	DIAO Z H, YUAN W B, DIAO C Y, et al.. Application of disease characteristics in crop disease identification: a review [J]. Jiangsu Agric. Sci., 2019, 47(5): 71-74.
3	陈雷,袁媛.基于深度迁移学习的农业病害图像识别[J].数据与计算发展前沿,2020,2(2):111-119.
	CHEN L, YUAN Y. Image recognition of agricultural diseases based on deep transfer learning [J]. Front Data Computing, 2020, 2(2): 111-119.
4	周惠汝,吴波明.深度学习在作物病害图像识别方面应用的研究进展[J].中国农业科技导报,2021,23(5): 61-68.
	ZHOU H R, WU B M. Research progress of deep learning application in crop disease image recognition [J]. J. Agric. Sci. Technol., 2021, 23(5): 61-68.
5	王佳盛,陈燕,曾泽钦,等.基于全卷积神经网络的荔枝表皮缺陷提取[J].华南农业大学学报,2018,39(6): 104-110.
	WANG J S, CHEN Y, ZHENG Z Q, et al.. Extraction of lichee epidermal defects based on full convolutional neural network [J]. J. South China Agric.Univ., 2018, 39(6): 104-110.
6	KARL R W, TAGHI K, DINGDING W. A survey of transfer learning [J]. J. Big Data, 2016,3(1): 37-43.
7	杨玉敏,廖育荣,林存宝,等.轻量化卷积神经网络目标检测算法综述[J].舰船电子工程,2021,41(4):31-36.
	YANG Y M, LIAO Y R, LIN C B, et al.. A review of target detection algorithms for lightweight convolutional neural networks [J]. Ship Electron. Eng., 2021, 41(4): 31-36.
8	史红栩,李修华,李民赞,等.基于深度学习的香蕉病害远程诊断系统[J].华南农业大学学报,2020,41(6): 92-99.
	SHI H X, LI X H, LI M Z, et al.. Remote diagnosis system of banana disease based on deep learning [J]. J. South China Agric. Univ., 2020, 41(6): 92-99.
9	李静,陈桂芬,安宇.基于优化卷积神经网络的玉米螟虫害图像识别[J].华南农业大学学报,2020,41(3):110-116.
	LI J, CHEN G F, AN Y. Image recognition of corn borer pests based on optimized convolutional neural network [J]. J. South China Agric. Univ., 2020, 41(3): 110-116.
10	龙满生,欧阳春娟,刘欢,等.基于卷积神经网络与迁移学习的油茶病害图像识别[J].农业工程学报,2018, 34(18):194-201.
	LONG M S, OUYANG C J, LIU H, et al.. Image recognition of camellia oil disease based on convolutional neural network and transfer learning [J]. Trans. Chin. Soc. Agric. Eng., 2018, 34(18): 194-201.
11	HOWARD A G, ZHU M, CHEN B, et al.. MobileNets: efficient convolutional neural networks for mobile vision applications [J]. 2017. 10(6): 187-191.
12	ALEX K, ILYA S, GEOFFREY E. H. ImageNet classification with deep convolutional neural networks [J]. Commun. ACM, 2017, 60(6):84-90.
13	CHOLLET F. Xception: deep learning with depthwise separable convolutions [J]. CoRR, 2017:1800-1807.
14	WILSON A C, ROELOFS R, STERN M, et al.. The marginal value of sdaptive gradient methods in machine learning [C/OL]// Proceedings of 31st Conference on Neural Information Processing Systems (NIPS 2017) [2023-06-15]. .
15	SASHANK J. R, SATYEN K, SANJIV K. On the convergence of adam and beyond [C/OL]// Proceedings of ICLR 2018 [2023-06-15]. .

病害类别 Disease category	原图像数量 Number of original images	扩充后数量 Number after expansion
枯萎病 Blight	538	1 093
锈蚀病 Common rust	286	818
灰叶斑病 Gray leaf spot	429	1 316
健康 Healthy	274	961

病害类别 Disease category	原图像数量 Number of original images	扩充后数量 Number after expansion
枯萎病 Blight	538	1 093
锈蚀病 Common rust	286	818
灰叶斑病 Gray leaf spot	429	1 316
健康 Healthy	274	961

输入 Input	模块 Operator	系数 Coefficient	输出通道Output channel	次数Times	步长Step
224²×3	Conv2d	—	32	1	2
112²×32	Bottleneck	1	16	1	1
112²×16	Bottleneck	6	24	2	2
56²×24	Bottleneck	6	32	3	2
28²×32	Bottleneck	6	64	4	2
14²×64	Bottleneck	6	96	3	1
14²×96	Bottleneck	6	160	3	2
7²×160	Bottleneck	6	320	1	1
7²×320	Conv2d	—	1 280	1	1
7²×1 280	Avgpool	—	—	1	—
1²×1 280	Conv2d	—	k	—	—

输入 Input	模块 Operator	系数 Coefficient	输出通道Output channel	次数Times	步长Step
224²×3	Conv2d	—	32	1	2
112²×32	Bottleneck	1	16	1	1
112²×16	Bottleneck	6	24	2	2
56²×24	Bottleneck	6	32	3	2
28²×32	Bottleneck	6	64	4	2
14²×64	Bottleneck	6	96	3	1
14²×96	Bottleneck	6	160	3	2
7²×160	Bottleneck	6	320	1	1
7²×320	Conv2d	—	1 280	1	1
7²×1 280	Avgpool	—	—	1	—
1²×1 280	Conv2d	—	k	—	—

模型 Model	模型大小 Size/Mb	准确率 Accuracy/%	耗时Time consumption/ms
VGG16	528	90.8	4.2
ResNet50	98	92.1	4.6
InceptionV3	92	93.7	6.9
Xception	88	94.5	8.1
MobileNetV2	14	90.1	3.8