[1]胡玲艳,周婷,刘艳,等.基于轻量级网络自适应特征提取的番茄病害识别[J].江苏农业学报,2022,38(03):696-705.[doi:doi:10.3969/j.issn.1000-4440.2022.03.015]
　HU Ling-yan,ZHOU Ting,LIU Yan,et al.Tomato disease recognition based on lightweight network auto-adaptive feature extraction[J].,2022,38(03):696-705.[doi:doi:10.3969/j.issn.1000-4440.2022.03.015]
点击复制

基于轻量级网络自适应特征提取的番茄病害识别()
分享到：

江苏农业学报[ISSN:1006-6977/CN:61-1281/TN]

卷:
38
期数:
2022年03期
页码:
696-705
栏目:
农业信息工程
出版日期:
2022-06-30

文章信息/Info

Title:
Tomato disease recognition based on lightweight network auto-adaptive feature extraction
作者:
胡玲艳; 周婷; 刘艳; 许巍; 盖荣丽; 李晓梅; 裴悦琨; 汪祖民
(大连大学信息工程学院，辽宁大连116622)
Author(s):
HU Ling-yan; ZHOU Ting; LIU Yan; XU Wei ; GAI Rong-li; LI Xiao-mei; PEI Yue-kun; WANG Zu-min
(School of Information Engineering， Dalian University， Dalian 116622, China)
关键词:
轻量级网络; 正形机制; 特征提取; 番茄; 病害识别
Keywords:
light weight network; correction mechanism; feature extraction; tomato; disease recognition
分类号:
TP391；S641.2
DOI:
doi:10.3969/j.issn.1000-4440.2022.03.015
文献标志码:
A
摘要:
为了实现番茄病害的精准识别，本研究提出一种轻量级网络自适应特征提取方法。该方法首先对图片进行正形处理，然后基于SqueezeNet模型构建轻量级网络模型GKFENet。GKFENet模型包含全局特征提取和关键特征提取2个模块，其中全局特征提取模块逐层提取番茄病害叶片的全局特征，关键特征提取模块通过学习评估出特征图各通道的重要程度，计算出权重值，最后将该值加权到原特征图上，从而实现病害关键特征的自适应提取。结果显示，正形机制有助于神经网络学习特征，本研究构建的GKFENet模型的平均识别准确率为97.90%，模型大小仅为2.64 MB，且在强噪声环境下，其识别准确率仍能保持在78.00%以上。GKFENet模型在训练过程中相对稳定，对8种番茄病害的识别准确率均超过96.00%。相比Bayes、KNN、LeNet、SqueezeNet、MobileNet模型，本研究构建的GKFENet模型的识别精度高，稳定性强且占用内存小，对于移动端未来的应用具有较高的实际价值。
Abstract:
To realize accurate recognition of tomato diseases, a lightweight network auto-adaptive feature extraction method was proposed. This method firstly performed a correction processing on the image. Then, based on the SqueezeNet model, a lightweight network model named global and key feature extraction network (GKFENet) was built. The GKFENet model included global feature extraction and key feature extraction modules. The global feature extraction module extracted the global features of tomato diseased leaves layer by layer. The key feature extraction module evaluated the importance of each channel in the feature images through learning, and calculated the weight value. Finally, the value was weighted on the original feature images to realize the adaptive extraction of the key features of the diseases. The results showed that the correction mechanism helped the neural network to learn features, the average identification accuracy of GKFENet model constructed in this research was 97.90%, and the model size was only 2.64 MB. In an environment with strong noise, the recognition accuracy of the model was still above 78.00%. GKFENet model was relatively stable during the training process, and the recognition accuracies of eight types of tomato diseases were all over 96.00%. Compared with Bayes, k-nearest neighbor (KNN), LeNet, SqueezeNet and MobileNet models, the GKFENet model constructed in this research has higher recognition accuracy, stronger stability and less memory. It has strong practical value in future mobile applications.

参考文献/References:

[1]TM P, PRANATHI A, SAIASHRITHA K, et al. Tomato leaf disease detection using convolutional neural networks: 2018 International conference on contemporary computing (IC3) [C]. Piscataway: IEEE Press, 2018.
[2]MIM T, SHEIKH M H, SHAMPA R A, et al. Leaves diseases detection of tomato using image processing: 2019 International conference on system modeling and advancement in research trends (SMART) [C]. Piscataway: IEEE Press, 2019.
[3]MOKHTAR U, ALI M A S, HASSANIEN A E, et al. Identifying two of tomatoes leaf viruses using support vector machine[M]. New Delhi: Springer Press, 2015:771-782.
[4]柴洋,王向东. 基于图像处理的温室大棚中番茄的病害识别[J]. 自动化技术与应用, 2013, 32(9): 83-89.
[5]JIANG D, LI F, YANG Y, et al. A tomato leaf diseases classification method based on deep learning: 2020 Chinese control and decision conference (CCDC) [C]. Piscataway: IEEE Press, 2020.
[6]RANGARAJAN A K, PURUSHOTHAMAN R, RAMESH A. Tomato crop disease classification using pre-trained deep learning algorithm[J]. Procedia Computer Science, 2018, 133: 1040-1047.
[7]王艳玲,张宏立,刘庆飞,等. 基于迁移学习的番茄叶片病害图像分类[J]. 中国农业大学学报, 2019, 24(6): 124-130.
[8]JIA S, JIA P, HU S, et al. Automatic detection of tomato diseases and pests based on leaf images: 2017 Chinese automation congress (CAC) [C]. Piscataway: IEEE Press, 2017.
[9]WU Q, CHEN Y, MENG J. DCGAN based data augmentation for tomato leaf disease identification[J]. IEEE Access, 2020,8: 98716-98728.
[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.
[11]李晓振,徐岩,吴作宏,等. 基于注意力神经网络的番茄叶部病害识别系统[J]. 江苏农业学报, 2020, 36(3): 561-568.
[12]MEERADEVI A K, RANJANA V, MUNDADA M R, et al. Design and development of efficient techniques for leaf disease detection using deep convolutional neural networks: 2020 IEEE international conference on distributed computing vlsi electrical circuits and robotics (DISCOVER) [C]. Piscataway: IEEE Press, 2020.
[13]方晨晨,石繁槐. 基于改进深度残差网络的番茄病害图像识别[J]. 计算机应用, 2020, 40(增刊1): 203-208.
[14]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.
[15]ELHASSOUNY A, SMARANDACHE F. Smart mobile application to recognize tomato leaf diseases using convolutional neural networks: 2019 International conference of computer science and renewable energies (ICCSRE) [C]. Piscataway: IEEE Press, 2019.
[16]AGARWAL M, GUPTA S K, BISWAS K K. Development of efficient CNN model for tomato crop disease identification [J]. Sustainable Computing: Informatics and Systems, 2020, 28（1）：100407.
[17]HU J, SHEN L, SUN G, et al. Squeeze-and-excitation networks: IEEE conference on computer vision and pattern recognition[C]. Piscataway: IEEE Press, 2017.
[18]HUGHES D P, SALATHE M. An open access repository of images on plant health to enable the development of mobile disease diagnostics[EB/OL]. (2016-04-04)
[2021-09-20]. https://arxiv.53yu.com/ftp/arxiv/papers/1511/1511.08060.
[19]刘子记,杜公福,牛玉,等. 番茄主要病害的发生与防治技术[J]. 长江蔬菜, 2019(19): 59-62.
[20]刘鹏鹏. 基于深度学习的番茄叶面型病虫害识别研究[D]. 南昌：南昌大学, 2020.
[21]CANNY J. A computational approach to edge detection[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 1986, 8(6): 679-698.
[22]宋余庆,谢熹,刘哲,等. 基于多层EESP深度学习模型的农作物病虫害识别方法[J]. 农业机械学报, 2020, 51(8): 196-202.
[23]IANDOLA F N, HAN S, MOSKEWICZ M W, et al. SqueezeNet：Alexnet-level accuracy with 50× fewer parameters and ＜0.5 MB model size[EB/OL]. (2016-11-04)
[2021-09-20].https://arxiv.org/abs/1602.07360v4.
[24]宋永嘉,刘宾,魏暄云,等. 大数据时代无线传感技术在精准农业中的应用进展[J].江苏农业科学,2021,49(8):31-37.
[25]李仁路,万书勤,康跃虎,等. 基于微灌工程设计成果数据的农田电子地图构建方法[J].排灌机械工程学报,2020,38(9):939-944.
[26]林娜,陈宏,赵健,等. 轻小型无人机遥感在精准农业中的应用及展望[J].江苏农业科学,2020,48(20):43-48.

备注/Memo

备注/Memo:
收稿日期：2021-09-20基金项目：国家自然科学基金青年基金项目(61601076)；大连市科技创新基金项目(2020JJ26SN058)作者简介：胡玲艳(1978-)，女，河北沧州人，博士，副教授，主要从事智慧农业、作物动态生长监测研究。(E-mail)hulingyan@dlu.edu.cn通讯作者：汪祖民，(E-mail)wangzumin@dlu.edu.cn
更新日期/Last Update: 2022-07-07