[1]肖宇,吴杰*,马驰.用于植物病虫害图像识别的数据增强方法[J].计算机技术与发展,2025,(03):210-214.[doi:10.20165/j.cnki.ISSN1673-629X.2024.0334]
　XIAO Yu,WU Jie*,MA Chi.A Data Augmentation Method for Image Recognition of Plant Pests and Diseases[J].,2025,(03):210-214.[doi:10.20165/j.cnki.ISSN1673-629X.2024.0334]
点击复制

用于植物病虫害图像识别的数据增强方法()

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

卷:
期数:
2025年03期
页码:
210-214
栏目:
新型计算应用系统
出版日期:
2025-03-10

文章信息/Info

Title:
A Data Augmentation Method for Image Recognition of Plant Pests and Diseases
文章编号:
1673-629X(2025)03-210-05
作者:
肖宇1; 吴杰1*; 马驰2
1. 辽宁科技大学 计算机与软件工程学院,辽宁 鞍山 114051;
2. 惠州学院 计算机科学与工程学院,广东 惠州 516007
Author(s):
XIAO Yu1; WU Jie1*; MA Chi2
1. School of Computer and Software Engineering,University of Science and Technology Liaoning,Anshan 114051,China;
2. School of Computer Scienceand Engineering,Huizhou University,Huizhou 516007,China
关键词:
数据增强; 类激活映射; 深度学习; 植物病虫害识别; GhostNet
Keywords:
data augmentation; class activation mapping; deep learning; plant pests and diseases recognition; GhostNet
分类号:
TP391
DOI:
10.20165/j.cnki.ISSN1673-629X.2024.0334
摘要:
在深度学习的植物病虫害图像识别领域,区域数据增强是提高模型泛化能力的关键策略。 该技术通过有选择性地移除图像的特定区域,促使模型更加专注于提取那些区分度较低的特征,进而增强了模型对新数据的适应性和识别能力。 所提出的 SaliencyBatchMix 数据增强方法利用类激活映射(CAM)计算语义百分比图(SPM)后,按 Batch 的维度选择具有代表性的图像区域,并将图像区域作为指示性补丁与目标图像混合,以引导模型学习更恰当的特征表示。 该方法可减少训练中裁剪区域的无意义像素,并且减少了标签噪声。 在 GhostNet 架构下的实验中,SaliencyBatchMix 分别在 IP102 和 Embrapa 数据集上实现了 72. 05% 和 96. 86% 的准确率。 对比于使用 CutMix 方法分别提升了 0. 62 百分点和 1 百分点。 通过对结果的对比和消融实验分析,验证了 SaliencyBatchMix 在提高模型泛化能力和准确率的有效性。
Abstract:
In the field of deep learning image recognition of plant pests and diseases,regional data augmentation is a key strategy to improve model generalization. It prompts models to focus on extracting less discriminative features by selectively removing image regions,thereby improving adaptability to new data. The proposed SaliencyBatchMix method utilizes Class Activation Mapping (CAM) to derive Semantic Percentage Maps ( SPM) after selecting representative regions based on batch dimensionality. It mixes indicative patches into target images to guide learning more suitable representations. This reduces meaningless cropped pixels during training and la-beling noise. Experiments using GhostNet show SaliencyBatchMix achieves 72. 05% and 96. 86% accuracy on IP102 and Embrapa,out-performing CutMix by 0. 62 percentage points and 1 percentage points, respectively. Results validation and ablation study findings corroborate SaliencyBatchMix’s effectiveness in boosting model generalization and accuracy.

相似文献/References:

[1]高友文,周本君,胡晓飞.基于数据增强的卷积神经网络图像识别研究[J].计算机技术与发展,2018,28(08):62.[doi:10.3969/ j. issn.1673-629X.2018.08.013]
　GAO You-wen,ZHOU Ben-jun,HU Xiao-fei.Research on Image Recognition of Convolution Neural Network Based on Data Enhancement[J].,2018,28(03):62.[doi:10.3969/ j. issn.1673-629X.2018.08.013]
[2]王海文,邱晓晖.一种基于生成式对抗网络的图像数据扩充方法[J].计算机技术与发展,2020,30(03):51.[doi:10. 3969 / j. issn. 1673-629X. 2020. 03. 010]
　WANG Hai-wen,QIU Xiao-hui.An Image Data Augmentation Method Based on Generative Adversarial Network[J].,2020,30(03):51.[doi:10. 3969 / j. issn. 1673-629X. 2020. 03. 010]
[3]龚 安,郭文婷.基于卷积神经网络的皮肤癌识别方法[J].计算机技术与发展,2020,30(10):167.[doi:10. 3969 / j. issn. 1673-629X. 2020. 10. 030]
　GONG An,GUO Wen-ting.Skin Cancer Image Classification Method Based on Convolutional Neural Network[J].,2020,30(03):167.[doi:10. 3969 / j. issn. 1673-629X. 2020. 10. 030]
[4]李天举,谢志峰,张侃弘,等.基于集成学习的烟草异常数据挖掘研究与应用[J].计算机技术与发展,2020,30(11):128.[doi:10. 3969 / j. issn. 1673-629X. 2020. 11. 024]
　LI Tian-ju,XIE Zhi-feng,ZHANG Kan-hong,et al.Study and Application of Tobacco Anomaly Data Mining Based on Ensemble Learning[J].,2020,30(03):128.[doi:10. 3969 / j. issn. 1673-629X. 2020. 11. 024]
[5]白瑜颖,刘宁钟,姜晓通.结合注意力混合裁剪的细粒度分类网络[J].计算机技术与发展,2021,31(10):38.[doi:10. 3969 / j. issn. 1673-629X. 2021. 10. 007]
　BAI Yu-ying,LIU Ning-zhong,JIANG Xiao-tong.Fine Grained Image Classification Network Combined with Attention CutMix[J].,2021,31(03):38.[doi:10. 3969 / j. issn. 1673-629X. 2021. 10. 007]
[6]凌永标,毛 峰,杨岚岚,等.基于混合注意力网络的安全工器具检测[J].计算机技术与发展,2022,32(06):209.[doi:10. 3969 / j. issn. 1673-629X. 2022. 06. 035]
　LING Yong-biao,MAO Feng,YANG Lan-lan*,et al.Safety Tools Detection Based on Hybrid Attention Network[J].,2022,32(03):209.[doi:10. 3969 / j. issn. 1673-629X. 2022. 06. 035]
[7]刘鸿瑜,袁国武.基于改进 YOLOv5s 的烟支外观缺陷检测方法[J].计算机技术与发展,2022,32(08):161.[doi:10. 3969 / j. issn. 1673-629X. 2022. 08. 026]
　LIU Hong-yu,YUAN Guo-wu.Detection of Cigarette Appearance Defects Based on Improved YOLOv5s[J].,2022,32(03):161.[doi:10. 3969 / j. issn. 1673-629X. 2022. 08. 026]
[8]黄志伦,刘 俊,郑 萌.改进 GAN 的光化性角化病图像数据增强方法[J].计算机技术与发展,2022,32(09):36.[doi:10. 3969 / j. issn. 1673-629X. 2022. 09. 006]
　HUANG Zhi-lun,LIU Jun,ZHENG Meng.An Improved Generation Adversarial Networks Method for Data Augmentation of Actinic Keratosis Image[J].,2022,32(03):36.[doi:10. 3969 / j. issn. 1673-629X. 2022. 09. 006]
[9]邓 辉,张 洁.基于改进的 ResNet50 网络的黑色素瘤分类方法[J].计算机技术与发展,2023,33(02):64.[doi:10. 3969 / j. issn. 1673-629X. 2023. 02. 010]
　DENG Hui,ZHANG Jie.Research on Skin Cancer Classification Method Based on Improved ResNet50[J].,2023,33(03):64.[doi:10. 3969 / j. issn. 1673-629X. 2023. 02. 010]
[10]高 贵,赵 阳,于舒娟,等.基于 GNN 的文本分类算法研究[J].计算机技术与发展,2023,33(05):138.[doi:10. 3969 / j. issn. 1673-629X. 2023. 05. 021]
　GAO Gui,ZHAO Yang,YU Shu-juan,et al.Research on Text Classification Algorithm Based on GNN[J].,2023,33(03):138.[doi:10. 3969 / j. issn. 1673-629X. 2023. 05. 021]

更新日期/Last Update: 2025-03-10