[1]白雪松,吴建平,景文超,等.基于改进残差网络的农作物病虫害检测研究[J].计算机技术与发展,2023,33(05):145-151.[doi:10. 3969 / j. issn. 1673-629X. 2023. 05. 022]
　BAI Xue-song,WU Jian-ping,JING Wen-chao,et al.Research on Detection of Crop Disease and Insect Pest Based on Improved Residual Network[J].,2023,33(05):145-151.[doi:10. 3969 / j. issn. 1673-629X. 2023. 05. 022]

点击复制

基于改进残差网络的农作物病虫害检测研究()

分享到：

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:

33

期数:

2023年05期

页码:

145-151

栏目:

人工智能

出版日期:

2023-05-10

文章信息/Info

Title:

Research on Detection of Crop Disease and Insect Pest Based on Improved Residual Network

文章编号:

1673-629X(2023)05-0145-07

作者:

白雪松¹ ; 吴建平¹; 2; 3 ; 景文超¹ ; 崔亚楠¹ ; 康小霖¹

1. 云南大学 信息学院,云南 昆明 650504;

2. 云南省电子计算中心,云南 昆明 650223;

3. 云南省高校数字媒体技术重点实验室,云南 昆明 650223

Author(s):

BAI Xue-song1 ; WU Jian-ping1; 2; 3 ; JING Wen-chao1 ; CUI Ya-nan1 ; KANG Xiao-lin1

1. School of Information Science & Engineering,Yunnan University,Kunming 650504,China;

2. Yunnan Provincial Electronic Computing Center,Kunming 650223,China;

3. Digital Media Technology Key Laboratory of Universities and Colleges in Yunnan Province,Kunming 650223,China

关键词:

农作物病虫害; 卷积神经网络; Res2NeXt50; 混合卷积; 标签平滑; 细粒度特征

Keywords:

crop disease and insect pest;  convolutional neural network;  Res2NeXt50;  mixed convolution;  label smoothing;  fine -grained featur

分类号:

TP312

DOI:

10. 3969 / j. issn. 1673-629X. 2023. 05. 022

摘要:

针对病虫害症状相似导致类间差异小、难以区分的问题,提出一种基于 Res2NeXt50 改进模型的农作物病虫害检测算法。 首先,在 Res2Net50 模型中进行分组卷积得到 Res2NeXt50 模型,提高

了模型在细粒度层面的特征提取能力。 然后,将 7×7 卷积换成新的混合卷积,提取局部和全局特征;使用高斯误差线性单元( Gaussian Error Linear Unit,GELU) 函数代替残差块中的修正线性单元

( Rectified Linear Unit,ReLU) 函数,提高鲁棒性;改进下采样来增强信息流通性;调整网络层数,以减少模型计算量。 其次,在训练中使用标签平滑( Label Smoothing) 和指数移动平均( Exponential Moving Average,EMA)来提高模型的泛化能力。 在重组的 AI Challenger 2018 农作物病虫害数据集上进行实验,结果表明改进模型的准确率高达98. 79% ,参数量为 18. 20M,FLOPs 为 3. 73G。 同时,该模型在 Plantvillage 和 Plant_leaves 数据集中分别达到了 99. 89% 和99. 23% 的准确率。 所提出的算法模型识别准确率高,泛化能力强,更符合实际应用需求。

Abstract:

Aiming at the problem of small differences and indistinguishability between classes due to similar symptoms of disease andinsect pest,a crop disease and insect pest detection algorithm based on the improved Res2NeXt50 model is proposed. Firstly,groupingconvolution is performed in the Res2Net50 model to obtain the Res2NeXt50 model,which improves the feature extraction capability ofthe model at the fine- grained level. Then,the 7×7　convolution is replaced with a new mixed convolution to extract local and globalfeatures. The Gaussian Error Linear Unit ( GELU) is used to replace the Rectified Linear Unit ( ReLU) in the residual block in order toimprove robustness. Downsampling is improved to enhance information flow,and the number of network layers is adjusted to reduce theamount of model computation. Secondly, Label Smoothing and Exponential Moving Average are used in training to improve thegeneralization ability of the model. Experiments were carried out on the restructured AI Challenger 2018 crop disease and insect pestdataset,which showed that the accuracy of the improved model was as high as 98. 79% ,the number of parameters was 18. 20M,and theFLOPs was 3. 73G. Meanwhile, the model achieves 99. 89% and 99. 23% accuracy in Plantvillage and Plant _ leaves datasets,　respectively. The proposed algorithm model has high recognition accuracy and strong generalization ability,which is more in line withpractical application requirements.

常用功能