双分支网络的苹果叶部病害识别

doi:10.3778/j.issn.1673-9418.2010013

计算机科学与探索 ›› 2022, Vol. 16 ›› Issue (4): 917-926.DOI: 10.3778/j.issn.1673-9418.2010013

双分支网络的苹果叶部病害识别

陆仲达¹^,⁺(), 张春达², 张佳奇², 王子菲², 许军华²

1.齐齐哈尔大学机电工程学院,黑龙江齐齐哈尔 161000
2.齐齐哈尔大学计算机与控制工程学院,黑龙江齐齐哈尔 161000

收稿日期:2020-10-09 修回日期:2021-01-11 出版日期:2022-04-01 发布日期:2021-01-25
通讯作者: + E-mail: luzhongda@163.com
作者简介:陆仲达（1970—）,男,黑龙江哈尔滨人,博士,教授,主要研究方向为模式识别与图像处理、机器人控制、非线性系统。
张春达（1998—）,男,河北故城人,硕士研究生,主要研究方向为图像识别、语义分割。
张佳奇（1996—）,男,黑龙江大兴安岭人,硕士研究生,主要研究方向为物联网应用、智能电表、非线性网络控制系统、神经网络算法。
王子菲（1996—）,女,黑龙江大庆人,硕士研究生,主要研究方向为网络控制系统、非线性控制、鲁棒控制。
许军华（1997—）,女,黑龙江黑河人,硕士研究生,主要研究方向为多智能体、智能电网。
基金资助:
黑龙江省省属高等学校基本科研业务费科研项目(135409602);研究生创新科研项目(YJSCX2020008)

Identification of Apple Leaf Disease Based on Dual Branch Network

LU Zhongda¹^,⁺(), ZHANG Chunda², ZHANG Jiaqi², WANG Zifei², XU Junhua²

1. School of Mechanical and Electrical Engineering, Qiqihar University, Qiqihar, Heilongjiang 161000, China
2. School of Computer and Control Engineering, Qiqihar University, Qiqihar, Heilongjiang 161000, China

Received:2020-10-09 Revised:2021-01-11 Online:2022-04-01 Published:2021-01-25
About author:LU Zhongda, born in 1970, Ph.D., professor. His research interests include pattern recognition and image processing, robot control and nonlinear system.
ZHANG Chunda, born in 1998, M.S. candi-date. His research interests include image identifi-cation and semantic segmentation.
ZHANG Jiaqi, born in 1996, M.S. candidate. His research interests include Internet of things appli-cation, intelligent electric meters, nonlinear network-ed control systems and neural network algorithm.
WANG Zifei, born in 1996, M.S. candidate. Her research interests include networked control sys-tems, nonlinear control and robust control.
XU Junhua, born in 1997, M.S. candidate. Her research interests include multi-agent system and smart grid.
Supported by:
Fundamental Research Funds in Heilongjiang Provincial Universities(135409602);Graduate Student Scientific Research Innovation Project(YJSCX2020008)

摘要/Abstract

摘要：

由于复杂背景环境和病斑相似性的影响,苹果叶部病害特征间存在细微的类间差异以及较大的类内差距,给苹果叶部病害识别造成极大困难。针对以上问题,提出了一种新型双分支网络的苹果叶部病害识别方法（DBNet）。DBNet的双分支网络结构由多尺度联合分支（MS）以及多维度注意力分支（DA）构成。首先多尺度联合分支通过不同类型卷积核和跨层连接融合不同尺度层级间的病害特征,用于缓解复杂背景环境带来的不利影响。同时多维度注意力分支通过融合宽、高、通道三个不同维度的注意力,使网络关注病斑间的微小差异,并随着网络层数的加深自动改变三个维度注意力所占比重,该分支用于缓解部分病斑相似性带来的不利影响。最终DBNet将双分支网络提取到的多尺度特征和多维度特征进行融合。并在苹果叶部病害数据集上,与AlexNet、VGG-16、ResNet-50、B-CNN等模型进行实验对比,结果显示所提方法能够有效地提升识别精度。

关键词: 苹果叶部病害, 双分支网络, 卷积神经网络（CNN）, 多尺度信息, 注意力机制

Abstract:

Due to the complex background environment and the similarity of disease spots, there are subtle inter-class differences and large intra-class differences in the pathological images of apple leaves, which causes great difficulties in the identification of apple leaves diseases. To solve this problem, a new dual branch network (DBNet) for apple leaf pathology identification is proposed. DBNet is composed of multi-scale branch (MS) and multi-dimensional attention branch (DA). MS branch fuses pathological features of different scales and levels through different types of convolution kernel and cross-layer connections, so as to alleviate the impact brought by complex background environment. At the same time, the DA branch makes the network pay attention to the small differences between disease spots by integrating the attention of three different dimensions, namely width, height and channel, and automatically changes the proportion of the attention of three dimensions with the deepening of the network layers. This branch is used to alleviate the adverse effects caused by the similarity of some disease spots. Finally, DBNet integrates multi-scale features and multi-dimensional features extracted from the dual-branch network. In addition, the experimental comparison with AlexNet, VGG-16, ResNet-50 and B-CNN on the apple leaf pathology dataset shows that the proposed method can effectively improve the recognition accuracy.

Key words: apple leaf disease, dual branch network, convolutional neural network (CNN), multi-scale information, attention mechanism

中图分类号:

TP391

陆仲达, 张春达, 张佳奇, 王子菲, 许军华. 双分支网络的苹果叶部病害识别[J]. 计算机科学与探索, 2022, 16(4): 917-926.

LU Zhongda, ZHANG Chunda, ZHANG Jiaqi, WANG Zifei, XU Junhua. Identification of Apple Leaf Disease Based on Dual Branch Network[J]. Journal of Frontiers of Computer Science and Technology, 2022, 16(4): 917-926.

图/表 18

图1 苹果叶部病害图像

Fig.1 Leaf disease diagram of apple

图2 DBNet网络结构图

Fig.2 DBNet network structure

图3 MS分支结构图

Fig.3 MS branch structure

图4 空洞卷积和非对称卷积

Fig.4 Dilated convolution and asymmetric convolutions

图5 跨层连接

Fig.5 Skip connections

图6 DA分支注意力模块结构图

Fig.6 DA branch attention module structure

图7 多维度注意力原理图

Fig.7 Diagram of multi-dimensional attention

表1 数据分布表

Table 1 Data distribution

苹果叶部病害	数据集	训练集	验证集
花叶病	4 875	3 412	1 463
褐斑病	5 655	3 958	1 697
锈病	5 694	3 985	1 709
灰斑病	4 810	3 367	1 443
斑点落叶病	5 343	3 740	1 603
总计	26 377	18 462	7 915

表2 实验环境配置表

Table 2 Experiment environment

实验环境配置项	参数值
CPU	Intel^®Xeon^®Gold 6271C CPU @2.60 GHz
显卡	NVIDIA Tesla V100 GPU 32 GB
内存	32 GB
磁盘	100 GB
深度学习框架	PaddlePaddle 1.8.4
操作系统	Ubuntu 16.04.1 LTS (64 bit)
其他工具	Python 3.7.1 CUDA10.1

表3 实验参数表

Table 3 Experiment parameters

参数名称	参数值
图像尺寸	224×224
批尺寸（batchsize）	60
学习率（learning rate）	0.001
训练周期（epoch）	30

图8 不同网络模型训练结果

Fig.8 Results of different network models

表4 网络模型准确率

Table 4 Accuracy of network models

网络模型	准确率（Accuracy）/%
AlexNet	86.898
VGG-16	92.255
ResNet-50	93.013
B-CNN	94.542
DBNet	97.662

图9 不同网络模型准确率变化对比

Fig.9 Comparison of accuracy changes of different network models

表5 网络复杂度

Table 5 Network complexity

网络模型	模型参数/10⁶	浮点运算次数/10⁹	CPU预测时间/ms
AlexNet	5.774 8	1.17	28
ResNet-50	23.518 3	8.19	179
VGG-16	138.357 6	30.84	501
B-CNN	291.586 8	61.69	1 138
DBNet	14.037 2	17.89	225

表6 MS消融实验

Table 6 MS ablation experiments

网络模型	准确率（Accuracy）/%
VGG	92.255
VGG+空洞卷积 $(d = 2)$	92.849
VGG+空洞卷积 $(d = 3)$	93.126
VGG+空洞卷积 $(d = 4)$	92.836
VGG+不对称卷积	93.493
VGG+空洞卷积 $(d = 2,3)$	94.011
VGG+不对称卷积+空洞卷积 $(d = 2)$	94.731
VGG+不对称卷积+空洞卷积 $(d = 3)$	94.883
VGG+不对称卷积+空洞卷积 $(d = 2,3)$	95.274

表6 MS消融实验

Table 6 MS ablation experiments

网络模型	准确率（Accuracy）/%
VGG	92.255
VGG+空洞卷积 $(d = 2)$	92.849
VGG+空洞卷积 $(d = 3)$	93.126
VGG+空洞卷积 $(d = 4)$	92.836
VGG+不对称卷积	93.493
VGG+空洞卷积 $(d = 2,3)$	94.011
VGG+不对称卷积+空洞卷积 $(d = 2)$	94.731
VGG+不对称卷积+空洞卷积 $(d = 3)$	94.883
VGG+不对称卷积+空洞卷积 $(d = 2,3)$	95.274

图10 四种结构准确率变化

Fig.10 Accuracy of four structures

表7 消融实验

Table 7 Ablation experiments %

苹果叶部病害	VGG+VGG	VGG+MS	VGG+DA	MS+DA
花叶病	99.658	100.000	100.000	100.000
褐斑病	96.994	98.055	98.232	98.939
锈病	93.680	94.499	94.675	95.846
灰斑病	95.357	95.564	97.228	97.643
斑点落叶病	93.637	94.448	95.757	97.005
五种苹果叶部病害	94.049	96.058	96.791	97.662

表8 模型对比实验

Table 8 Contrast experiment of models

网络模型	准确率（Accuracy）/%
文献[18]	94.819
文献[19]	95.250
文献[20]	96.222
文献[21]	96.412
DBNet	97.662

参考文献 21

[1]	张善文, 张晴晴, 李萍. 基于改进深度卷积神经网络的苹果病害识别[J]. 林业工程学报, 2019, 4(4):107-112.
	ZHANG S W, ZHANG Q Q, LI P. Apple disease identifi-cation based on improved deep convolutional neural net-work[J]. Journal of Forestry Engineering, 2019, 4(4):107-112.
[2]	王冠, 王建新, 孙钰. 面向边缘计算的轻量级植物病害识别模型[J]. 浙江农林大学学报, 2020, 37(5):978-985.
	WANG G, WANG J X, SUN J. Lightweight plant disease recognition model for edge computing[J]. Journal of Zhe-jiang A&F University, 2020, 37(5):978-985.
[3]	TIAN Y, ZHAO C J, LU S L, et al. Multiple classifier com-bination for recognition of wheat leaf diseases[J]. Intelli-gent Automation & Soft Computing, 2011, 17(5):519-529.
[4]	HIARY H A, AHMAD S B, REYALAT M, et al. Fast and accurate setection and classification of plant diseases[J]. International Journal of Computer Applications, 2011, 17(1):31-38.
[5]	师韵, 黄文准, 张善文. 基于二维子空间的苹果病害识别方法[J]. 计算机工程与应用, 2017, 53(22):180-184.
	SHI Y, HUANG W Z, ZHANG S W. Apple disease recog-nition based on two-dimensionality subspace learning[J]. Computer Engineering and Applications, 2017, 53(22):180-184.
[6]	孙素云. 基于图像处理和支持向量机的苹果树叶部病害的分类研究[D]. 西安: 西安科技大学, 2017.
	SUN S Y. Classification of apple leaf diseases based on image processing and support vector machine[D]. Xi’an: Xi’an University of Science and Technology, 2017.
[7]	KRIZHEVSKY A, SUTSKEVER I, HINTON G E. ImageNet classification with deep convolutional neural networks[J]. Communications of the ACM, 2017, 60(6):84-90. DOI URL
[8]	SZEGEDY C, LIU W, JIA Y, et al. Going deeper with con-volutions[C]// Proceedings of the 2015 IEEE Conference on Computer Vision and Pattern Recognition, Boston, Jun 7-12, 2015. Washington: IEEE Computer Society, 2015: 1-9.
[9]	张建华, 孔繁涛, 吴建寨, 等. 基于改进VGG卷积神经网络的棉花病害识别模型[J]. 中国农业大学学报, 2018, 23(11):161-171.
	ZHANG J H, KONG F T, WU J Z, et al. Cotton disease identification model based on improved VGG convolution neural network[J]. Journal of China Agricultural University, 2018, 23(11):161-171.
[10]	胡志伟, 杨华, 黄济民, 等. 基于注意力残差机制的细粒度番茄病害识别[J]. 华南农业大学学报, 2019, 40(6):124-132.
	HU Z W, YANG H, HUANG J M, et al. Fine-grained tomato disease recognition based on attention residual mechanism[J]. Journal of South China Agricultural University, 2019, 40(6):124-132.
[11]	王昌龙, 张远东, 缪宏, 等. 双通道卷积神经网络在南瓜病害识别上的应用[J]. 计算机工程与应用, 2021, 57(5):183-189.
	WANG C L, ZHANG Y D, MIAO H, et al. Application of double channel convolutional neural network in pumpkin diseases identification[J]. Computer Engineering and App-lications, 2021, 57(5):183-189.
[12]	LIN T Y, ROYCHOWDHURY A, MAJI S, et al. Bilinear CNN models for fine-grained visual recognition[C]// Procee-dings of the 2015 IEEE International Conference on Com-puter Vision, Santiago, Dec 7-13, 2015. Washington: IEEE Computer Society, 2015: 1449-1457.
[13]	MCNEELY-WHITE D G, BEVERIDGE J R, DRAPER B A. Inception and ResNet features are (almost) equivalent[J]. Cognitive Systems Research, 2020, 59:312-318. DOI URL
[14]	HU J, SHEN L, ALBANIE S, et al. Squeeze-and-excitation networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020, 42(8):2011-2023. DOI URL
[15]	PENG J, CHEN Y H, LIU B, et al. Real-time detection of apple leaf diseases using deep learning approach based on improved convolutional neural networks[J]. IEEE Access, 2019, 7:59069-59080. DOI URL
[16]	CHEN L C, PAPANDREOU G, KOKKINOS I, et al. Deep-Lab: semantic image segmentation with deep convolu-tional nets, atrous convolution, and fully connected CRFs[J]. IEEE Transactions on Pattern Analysis and Machine Intelli-gence, 2018, 40(4):834-848.
[17]	DING X H, GUO Y C, DING G G, et al. ACNet: streng-thening the kernel skeletons for powerful CNN via asymme-tric convolution blocks[C]// Proceedings of the 2019 IEEE International Conference on Computer Vision, Seoul, Oct 27-Nov 2, 2019. Washington: IEEE Computer Society, 2019: 1911-1920.
[18]	JENIFA A, RAMALAKSHMI R, RAMACHANDRAN V. Cotton leaf disease classification using deep convolution neural network for sustainable cotton production[C]// Pro-ceedings of the 2019 IEEE International Conference on Clean Energy and Energy Efficient Electronics Circuit for Sustainable Development, Krishnankoil, Dec 18-20, 2019. Washington: IEEE Computer Society, 2019: 1-3.
[19]	WAHEED A, GOYAL M, GUPTA D, et al. An optimized dense convolutional neural network model for disease re-cognition and classification in corn leaf[J]. Computers and Electronics in Agriculture, 2020, 175:105456. DOI URL
[20]	SYARIEF M, SETIAWAN W. Convolutional neural network for maize leaf disease image classification[J]. TELKOMNIKA (Telecommunication Computing Electronics and Control), 2020, 18(3):1376-1381. DOI URL
[21]	KHAN A I, QUADRI S M K, BANDAY S. Deep learning for apple diseases: classification and identification[J]. Inter-national Journal of Computational Intelligence Studies, 2021, 10(1):1-12.

双分支网络的苹果叶部病害识别

Identification of Apple Leaf Disease Based on Dual Branch Network

RichHTML

PDF

可视化

摘要/Abstract

引用本文

使用本文

图/表 18

参考文献 21

相关文章 15

编辑推荐

Metrics

[1]	杨知桥, 张莹, 王新杰, 张东波, 王玉. 改进U型网络在视网膜病变检测中的应用研究[J]. 计算机科学与探索, 2022, 16(8): 1877-1884.
[2]	安凤平, 李晓薇, 曹翔. 权重初始化-滑动窗口CNN的医学图像分类[J]. 计算机科学与探索, 2022, 16(8): 1885-1897.
[3]	夏鸿斌, 肖奕飞, 刘渊. 融合自注意力机制的长文本生成对抗网络模型[J]. 计算机科学与探索, 2022, 16(7): 1603-1610.
[4]	李运寰, 闻继伟, 彭力. 高帧率的轻量级孪生网络目标跟踪[J]. 计算机科学与探索, 2022, 16(6): 1405-1416.
[5]	张雁操, 赵宇海, 史岚. 融合图注意力的多特征链接预测算法[J]. 计算机科学与探索, 2022, 16(5): 1096-1106.
[6]	程卫月, 张雪琴, 林克正, 李骜. 融合全局与局部特征的深度卷积神经网络算法[J]. 计算机科学与探索, 2022, 16(5): 1146-1154.
[7]	童敢, 黄立波. Winograd快速卷积相关研究综述[J]. 计算机科学与探索, 2022, 16(5): 959-971.
[8]	赵鹏飞, 谢林柏, 彭力. 融合注意力机制的深层次小目标检测算法[J]. 计算机科学与探索, 2022, 16(4): 927-937.
[9]	裴利沈, 赵雪专. 群体行为识别深度学习方法研究综述[J]. 计算机科学与探索, 2022, 16(4): 775-790.
[10]	黄思远, 赵宇海, 梁燚铭. 融合图嵌入和注意力机制的代码搜索[J]. 计算机科学与探索, 2022, 16(4): 844-854.
[11]	王燕妮, 余丽仙. 注意力与多尺度有效融合的SSD目标检测算法[J]. 计算机科学与探索, 2022, 16(2): 438-447.
[12]	蒋光峰, 胡鹏程, 叶桦, 仰燕兰. 基于重构误差的同构图分类模型[J]. 计算机科学与探索, 2022, 16(1): 185-193.
[13]	李志欣, 陈圣嘉, 周韬, 马慧芳. 协同级联网络和对抗网络的目标检测[J]. 计算机科学与探索, 2022, 16(1): 217-230.
[14]	蔡明昕, 孙晶, 王斌. 多角度语义轨迹相似度计算模型[J]. 计算机科学与探索, 2021, 15(9): 1632-1640.
[15]	陈俊芬, 张明, 赵佳成, 谢博鋆, 李艳. 结合降噪和自注意力的深度聚类算法[J]. 计算机科学与探索, 2021, 15(9): 1717-1727.