基于坐标注意力机制和残差网络的水稻叶片病虫害识别

廖媛珺; 杨乐; 邵鹏; 余小云

doi:10.19303/j.issn.1008-0384.2023.10.011

基于坐标注意力机制和残差网络的水稻叶片病虫害识别

doi: 10.19303/j.issn.1008-0384.2023.10.011

廖媛珺^1,,
杨乐^{1, 2, ,},
邵鹏¹,
余小云¹

1.
江西农业大学计算机与信息工程学院，江西　南昌　330045
2.
江西省高等学校农业信息技术重点实验室，江西　南昌　330045

基金项目: 国家自然科学基金项目（61862032）；江西省自然科学基金项目（20202BABL202034）

详细信息

作者简介:
廖媛珺（2001 —），女，主要从事农业信息技术研究，E-mail：2426247587@qq.com

通讯作者:
杨乐（1979 —），男，副教授，硕士生导师，主要从事农业信息技术研究，E-mail：jxnzhyangle@163.com

中图分类号: S435
计量
- 文章访问数: 279
- HTML全文浏览量: 189
- PDF下载量: 58
- 被引次数: 0
出版历程
- 收稿日期: 2023-02-08
- 修回日期: 2023-03-10
- 网络出版日期: 2023-09-19
- 刊出日期: 2023-10-28

Improved Identification of Leaf Diseases and Pest Infestations on Rice by Means of Coordinate Attention Mechanism-based Residual Network

LIAO Yuanjun^1
,,
YANG Le^{1, 2
, ,},
SHAO Peng¹,
YU Xiaoyun¹

1.
College of Computer Information and Engineering, Jiangxi Agricultural University, Nanchang, Jiangxi　330045, China
2.
Key Laboratory of Information Technology in Agriculture for College and Universities of Jiangxi Province, Jiangxi Agriculture University, Nanchang, Jiangxi　330045, China

摘要

摘要: 目的针对在自然条件下水稻叶片病虫害的识别效率不高、准确率较低的问题，探索基于ResNet深度学习网络的水稻叶片病虫害识别模型（ResNet50-CA）。方法在ResNet-50的残差卷积模块下引入坐标注意力机制（CA），采用 LeakyReLU 激活函数替代 ReLU 激活函数，使用3个3×3的卷积核替换ResNet-50模型首层卷积层中的7×7卷积核。结果在使用传统卷积神经网络进行水稻叶片病虫害研究发现，ResNet-50能够较好地平衡识别准确率和模型复杂度之间的关系，因此选择在ResNet-50网络模型的基础上加以改进。使用改进后的网络通过微调参数进行水稻叶片病虫害对比性能试验，研究发现在批量样本数为16和学习率为0.0001时，ResNet50-CA获得最高的识别准确率（99.21%），优于传统的深度学习算法。结论改进后的网络能够提取出水稻病虫害更加细微的特征信息，从而取得更高的识别准确率，为水稻叶片病虫害识别提供新思路和方法。
- 深度学习 /
- ResNet50 /
- 水稻病虫害识别 /
- 坐标注意力机制
Abstract: Objective A new deep learning network was designed to improve the often-inaccurate identification of diseases and pest infestations on rice. Method The coordinate attention mechanism (CA) was introduced under the residual convolution block of RestNet-50 using the LeakyRelu activation function to replace the Relu activation function as well as the three 3×3 convolution kernels to replace the original 7×7 convolution kernel under the first convolution layer. Result The newly designed ResNet-50-CA effectively balanced the detection accuracy and model simplicity the original method lacked. The improved model was further fine-tuned with experiments to achieve a much-improved detection accuracy of 99.21% in identifying the diseases and infestations on a batch of 16 specimens with a learning rate of 0.0001. Conclusion The superior deep learning algorithm of the current ResNet50-CA system extracted more detailed and accurate information on the diseases and infestations than did the previous model. It could be applied for field and/or clinic diagnosis on rice plants.
- Deep learning network /
- ResNet50 /
- rice leaf diseases and pest infestations /
- coordinate attention mechanism

HTML全文

图 1 水稻叶片病虫害图像

Figure 1. Photos of 8 leaf diseases and pest infestations on rice plants

下载: 全尺寸图片幻灯片

图 2 坐标注意力的结构

Figure 2. Structure of coordinate attention

下载: 全尺寸图片幻灯片

图 3 残差网络的结构

Figure 3. Structure of residual network

下载: 全尺寸图片幻灯片

图 4 传统ResNet-50结构图

Figure 4. Structure of conventional ResNet 50

下载: 全尺寸图片幻灯片

图 5 relu激活函数的图像

Figure 5. Graphic Relu activation function

下载: 全尺寸图片幻灯片

图 6 LeakyRelu激活函数的图像

Figure 6. Graphic leakyRelu activation function

下载: 全尺寸图片幻灯片

图 7 ResNet50-CA的结构

Figure 7. Structure of ResNet50-CA

下载: 全尺寸图片幻灯片

图 8 训练集和测试集的准确率变化曲线

Figure 8. Accuracy of testing and validation datasets

下载: 全尺寸图片幻灯片

图 9 训练集和测试集的损失变化曲线

Figure 9. Losses of testing and validation datasets

下载: 全尺寸图片幻灯片

图 10 5种batchsize的准确率

Figure 10. Comparative accuracy of 5 batchsize

下载: 全尺寸图片幻灯片

图 11 5种batchsize的损失值

Figure 11. Comparative losses of 5 batchsize

下载: 全尺寸图片幻灯片

图 12 4种学习率的准确率对比图

Figure 12. Accuracy under 4 learning rates

下载: 全尺寸图片幻灯片

图 13 学习率的损失值对比图

Figure 13. Comparative losses of 4 learning rates

下载: 全尺寸图片幻灯片

表 1 水稻叶片病虫害的数据集

Table 1. Collected data on leaf diseases and pest infestations on rice plants

病虫害种类 The type of leaf pests and diseases	初始数据集 Initial dataset	数据增强后的数据集 The dataset after expansion
白叶枯病 Bacterial leaf smut	200	1279
褐斑病 Brown spot	200	1279
叶黑穗病 Leaf Smut	200	1279
纹枯病 Stiae blight	93	1279
干尖线虫病 Dry tip worm disease	76	1279
细菌性条斑病 Bacterial leaf streak	114	1279
赤枯病 Red blight	80	1279
稻瘟病 Leaf blast	123	1279

下载: 导出CSV

表 2 ResNet50-CA模型的水稻叶片病虫害数据集测试指标

Table 2. ResNet50-CA indicators on leaf diseases and pest infestations of rice

病害 Diseases	精确率 Accuracy/%	召回率 Recall/%	特异度 Specificity/%	F1/%
白叶枯病 Bacterial leaf smut	100.0	100.0	100.0	100.0
褐斑病 Brown spot	99.6	100.0	99.9	99.8
叶黑穗病 Leaf smut	99.6	99.6	99.9	99.6
纹枯病 Stiae blight	100.0	97.1	100.0	98.5
干尖线虫病 Dry tip worm disease	96.8	100.0	99.5	98.4
细菌性条斑病 Bacterial leaf streak	99.6	99.6	99.9	99.6
赤枯病 Red blight	99.6	98.7	99.9	99.2
稻瘟病 Leaf blast	98.7	98.7	99.8	98.7

下载: 导出CSV

表 3 不同神经网络的试验结果

Table 3. Experimental results of convolutional neural networks

模型名称 Model name	硬件性能 FLOPs/GMac	参数 Parameters/M	准确率 Accuracy/%
VGG13	11.27	65.00	96.82
VGG16	15.44	70.40	97.69
MobileNet	0.32	2.23	91.26
ResNet34	3.68	21.29	95.43
ResNet50	4.12	23.52	96.44
ResNet101	7.85	42.52	96.47
ResNet50-CA	4.99	25.51	99.21

下载: 导出CSV

表 4 不同数据集的对比结果

Table 4. Experimental results on plant datasets

模型名称 Model name	种类 Species	病虫害名称 Name of diseases	精确率 Precision/%	召回率 Recall/%	特异度 Specificity/%	F1/%	准确率 Accuracy/%
ResNet50	苹果 Apple	黑腐病 Black rot	100.0	99.5	100.0	99.7	99.78
		健康 Healthy	99.7	99.7	99.8	99.7
		赤锈病 Red rust	100.0	100.0	100.0	100.0
		赤霉病 Scab	99.5	100.0	99.9	99.7
	葡萄 Grape	黑麻疹病 Black measles	100.0	98.9	100.0	99.5	99.67
		黑腐病 Black rot	99.2	100.0	99.7	99.6
		叶枯病 Leaf blight	100.0	100.0	100.0	100.0
		健康 Healthy	99.5	100.0	99.9	99.7
	玉米 Corn	普通锈病 Common rust	99.6	100.0	99.8	99.8	98.6
		叶枯病 Leaf blight	99.0	95.5	99.7	97.2
		健康 Healthy	99.6	100.0	99.8	99.8
		叶斑病 Leaf spot	96.1	98.5	98.8	97.3
ResNet50-CA	苹果 Apple	黑腐病 Black rot	100.0	99.5	100.0	99.7	99.8
		健康 Healthy	99.7	100.0	99.8	99.8
		赤锈病 Red rust	100.0	100.0	100.0	100.0
		赤霉病 Scab	100.0	100.0	100.0	100.0
	葡萄 Grape	黑麻疹病 Black measles	99.6	100.0	99.8	99.8	99.8
		黑腐病 Black rot	100.0	99.6	100.0	99.8
		叶枯病 Leaf blight	100.0	100.0	100.0	100.0
		健康 Healthy	100.0	100.0	100.0	100.0
	玉米 Corn	普通锈病 Common rust	100.0	100.0	100.0	100.0	99.0
		叶枯病 Leaf blight	99.5	97.0	99.9	98.2
		健康 Healthy	100.0	99.6	100.0	99.8
		叶斑病 Leaf spot	99.6	99.5	99.0	98.0

下载: 导出CSV

参考文献(20)

[1]	ZHOU J, LI J X, WANG C S, et al. Crop disease identification and interpretation method based on multimodal deep learning [J]. Computers and Electronics in Agriculture, 2021, 189: 106408. doi: 10.1016/j.compag.2021.106408
[2]	王超, 王春圻, 刘金明. 基于深度学习的玉米叶片病害识别方法研究 [J]. 现代农业研究, 2022, 28(6):102−106. WANG C, WANG C Q, LIU J M. Identification of maize leaf diseases based on deep learning [J]. Modern Agriculture Research, 2022, 28(6): 102−106.(in Chinese)
[3]	JIANG F, LU Y, CHEN Y, et al. Image recognition of four rice leaf diseases based on deep learning and support vector machine [J]. Computers and Electronics in Agriculture, 2020, 179: 105824. doi: 10.1016/j.compag.2020.105824
[4]	BRAHIMI M, BOUKHALFA K, MOUSSAOUI A. Deep learning for tomato diseases: Classification and symptoms visualization [J]. Applied Artificial Intelligence, 2017, 31(4): 299−315. doi: 10.1080/08839514.2017.1315516
[5]	赵恒谦, 杨屹峰, 刘泽龙, 等. 农作物叶片病害迁移学习分步识别方法 [J]. 测绘通报, 2021(7):34−38. ZHAO H Q, YANG Y F, LIU Z L, et al. Step-by-step identification method of crop leaf diseases based on transfer learning [J]. Bulletin of Surveying and Mapping, 2021(7): 34−38.(in Chinese)
[6]	KAWASAKI Y, UGA H, KAGIWADA S, et al. Basic study of automated diagnosis of viral plant diseases using convolutional neural networks[M]//Advances in Visual Computing. Cham: Springer International Publishing, 2015: 638－645.
[7]	LU Y Z, YOUNG S. A survey of public datasets for computer vision tasks in precision agriculture [J]. Computers and Electronics in Agriculture, 2020, 178: 105760. doi: 10.1016/j.compag.2020.105760
[8]	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). Nashville, TN, USA. IEEE, 2021: 13708－13717.
[9]	王志强, 于雪莹, 杨晓婧, 等.基于WGAN和MCA-MobileNet的番茄叶片病害识别[J]. 农业机械学报, 2023, 54（5）: 244－252. WANG Z Q, YU X Y, YANG X J, et al. Tomato Leaf Diseases Recognition Based on WGAN and MCA - MobileNet[J]. 2023, 54(5): 244－251.
[10]	HE K, ZHANG X, REN S, et al. Deep residual learning for image recognition[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2016: 770−778.
[11]	XU B, WANG N Y, CHEN T Q, et al. Empirical evaluation of rectified activations in convolutional network[EB/OL]. 2015: arXiv: 1505.00853.https://arxiv.org/abs/1505.00853
[12]	李键红, 吴亚榕. 基于多任务联合稀疏表示的水稻叶片病害自动识别算法 [J]. 仲恺农业工程学院学报, 2017, 30(2):40−44. doi: 10.3969/j.issn.1674-5663.2017.02.007 LI J H, WU Y R. Automatic recognition algorithm for rice leaf diseases based on multi-task joint sparse representation [J]. Journal of Zhongkai University of Agriculture and Engineering, 2017, 30(2): 40−44.(in Chinese) doi: 10.3969/j.issn.1674-5663.2017.02.007
[13]	JIANG Z C, DONG Z X, JIANG W P, et al. Recognition of rice leaf diseases and wheat leaf diseases based on multi-task deep transfer learning [J]. Computers and Electronics in Agriculture, 2021, 186: 106184. doi: 10.1016/j.compag.2021.106184
[14]	刘立波, 周国民. 基于多层感知神经网络的水稻叶瘟病识别方法 [J]. 农业工程学报, 2009, 25(S2):213−217. LIU L B, ZHOU G M. Identification method of rice leaf blast using multilayer perception neural network [J]. Transactions of the Chinese Society of Agricultural Engineering, 2009, 25(S2): 213−217.(in Chinese)
[15]	MNIH V, HEESS N, GRAVES A, et al. Recurrent models of visual attention[C]//In Advances in neural information processing systems. 2014: 2204－2212.
[16]	XU K, BA J L, KIROS R, et al. Show, attend and tell: Neural image caption generation with visual attention[C]//Proceedings of the 32nd International Conference on International Conference on Machine Learning - Volume 37. July 6 - 11, 2015, Lille, France. New York: ACM, 2015: 2048－2057.
[17]	HU J, SHEN L, SUN G. Squeeze-and-excitation networks[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Salt Lake City, UT, USA. IEEE, 2018: 7132－7141.
[18]	WOO S, PARK J, LEE J Y, et al. CBAM: convolutional block attention module[M]//Computer Vision – ECCV 2018. Cham: Springer International Publishing, 2018: 3－19.
[19]	WANG H Y, ZHU Y K, GREEN B, et al. Axial-DeepLab: Stand-alone axial-attention for panoptic segmentation[EB/OL]. 2020: arXiv: 2003.07853. https://arxiv.org/abs/2003.07853
[20]	ZHAO H S, SHI J P, QI X J, et al. Pyramid scene parsing network[J]. IEEE Computer Society, 2016: 6230－6239.