[1]雷竣杰,周保平.基于改进DCGAN的棉叶螨为害图像数据增强方法[J].江苏农业学报,2025,(05):916-926.[doi:doi:10.3969/j.issn.1000-4440.2025.05.010]
　LEI Junjie,ZHOU Baoping.A data augmentation method for cotton leaf mite damage images based on improved DCGAN[J].,2025,(05):916-926.[doi:doi:10.3969/j.issn.1000-4440.2025.05.010]
点击复制

基于改进DCGAN的棉叶螨为害图像数据增强方法()
分享到：

江苏农业学报[ISSN:1006-6977/CN:61-1281/TN]

卷:
期数:
2025年05期
页码:
916-926
栏目:
农业信息工程
出版日期:
2025-05-31

文章信息/Info

Title:
A data augmentation method for cotton leaf mite damage images based on improved DCGAN
作者:
雷竣杰1; 2; 周保平1; 2
(1.塔里木大学信息工程学院,新疆阿拉尔843300；2.塔里木绿洲农业教育部重点实验室,新疆阿拉尔843300)
Author(s):
LEI Junjie1; 2; ZHOU Baoping1; 2
(1.College of Information Engineering, Tarim University, Alaer 843300, China;2.Key Laboratory of Tarim Oasis Agriculture, Ministry of Education, Alaer 843300, China)
关键词:
棉叶螨; 为害程度; 深度卷积生成对抗网络（DCGAN）; 图像数据增强
Keywords:
cotton leaf mite; damage degree; deep convolutional generative adversarial network (DCGAN); image data augmentation
分类号:
TP391.4
DOI:
doi:10.3969/j.issn.1000-4440.2025.05.010
文献标志码:
A
摘要:
为解决棉叶螨不同为害程度图像样本量不足和类别不平衡的问题,降低数据采集成本,并提高生成对抗网络生成图像的质量和多样性,本研究提出了一种基于改进DCGAN模型的棉叶螨为害图像数据增强方法。在原始模型的基础上,引入类别标签,使模型能够针对不同等级的棉叶螨为害图像进行针对性生成,有效解决类别不平衡问题；其次,将传统的直连结构替换为残差结构,增强模型对复杂映射关系的学习能力,避免梯度消失问题,提升生成图像的质量；接着,在卷积层中嵌入卷积注意力模块(CBAM),强化模型对棉叶螨为害图像关键特征的提取能力,进一步提高生成图像的质量和多样性；最后,采用带有梯度惩罚的Wasserstein距离作为损失函数,避免模式崩溃的问题,增强模型的训练稳定性。改进后的DCGAN模型在训练稳定性和生成图像质量方面均优于原始模型,其生成图像的Inception score（IS，8.51)、Fréchet inception distance（FID， 150.12)、Kernel inception distance（KID， 0.06)和结构相似性指数（SSIM， 0.82)均高于其他经典数据增强模型生成的图像。以改进的DCGAN模型生成的图像构建训练集训练棉叶螨为害图像分级模型——DenseNet-121模型,结果表明,基于改进的DCGAN模型生成的数据集训练的DenseNet-121模型平均分级准确率达88.02%,高于基于传统增强方法和其他模型生成的数据集训练的DenseNet-121模型。本研究为农业病虫害智能监测提供了技术支持。
Abstract:
To address the insufficient and imbalanced sample sizes of cotton leaf mite damage images at different severity levels, reduce data collection costs, and enhance the quality and diversity of images generated by generative adversarial networks, this study proposed an improved DCGAN-based data augmentation method for cotton leaf mite damage images. Based on the original model, category labels were introduced to enable targeted generation of images for different damage levels, effectively resolving the issue of class imbalance. The traditional direct connection structure was replaced with a residual structure to enhance the model’s ability to learn complex mapping relationships, avoid gradient vanishing problems, and improve the quality of generated images. Additionally, the convolutional block attention module (CBAM) was embedded in the convolutional layers to strengthen the model’s capacity to extract key features of cotton leaf mite damage images, further enhancing the quality and diversity of generated images. Lastly, the Wasserstein distance with gradient penalty was employed as the loss function, avoiding the problem of mode collapse and enhancing the training stability of the model. The improved DCGAN model outperformed the original model in terms of training stability and image quality. Its generated images achieved higher inception score (IS, 8.51), fréchet inception distance (FID, 150.12), kernel inception distance (KID, 0.06), and structural similarity index measure (SSIM, 0.82) than those generated by other classic data augmentation models. When training the DenseNet-121 model with the dataset generated by the improved DCGAN model, the average classification accuracy reached 88.02%, which was higher than that of DenseNet-121 models trained with datasets generated by traditional augmentation methods and other models. This study provides technical support for intelligent monitoring of agricultural pests and diseases.

参考文献/References:

[1]YANG Z N, TANG J J, YU M, et al. Sustainable cotton production through increased competitiveness:analysis of comparative advantage and influencing factors of cotton production in Xinjiang,China[J]. Agronomy,2022,12(10):2239.
[2]甄丛爱,买合吐木古力·艾孜不拉,张帅,等. 棉叶螨的抗药性现状与治理策略[J]. 植物保护学报,2023,50(3):559-577.
[3]张龚,周保平,王昱,等. 基于迁移学习和改进残差网络的棉花叶螨为害等级识别[J]. 浙江农业学报,2023,35(7):1729-1739.
[4]杨丽丽,张大卫,罗君,等. 基于SVM和AdaBoost的棉叶螨危害等级识别[J]. 农业机械学报,2019,50(2):14-20.
[5]何青海,马本学,瞿端阳,等. 基于机器视觉棉叶螨自动监测与分级方法研究[J]. 农机化研究,2013,35(4):152-155.
[6]章展熠,张宝荃,王周立,等. 多茶类CNN图像识别的数据增强优化及类激活映射量化评价[J]. 茶叶科学,2023,43(3):411-423.
[7]BANSAL M A, SHARMA D R, KATHURIA D M. A systematic review on data scarcity problem in deep learning:solution and applications[J]. ACM Computing Surveys (Csur),2022,54(10):1-29.
[8]周思聪,向峰,李红军,等. 基于自编码器GAN数据增强的工业小目标缺陷检测[J]. 现代制造工程,2025,(2):101-108.
[9]GOODFELLOW I, POUGET-ABADIE J, MIRZA M, et al. Generative adversarial nets[C]//NIPS. Proceedings of the 27th international conference on neural information processing systems. Cambridge：MIT Press,2014.
[10]SHARMA V, TRIPATHI A K, DAGA P, et al. ClGanNet:a novel method for maize leaf disease identification using ClGan and deep CNN[J]. Signal Processing:Image Communication,2024,120:117074.
[11]WANG F Y, RAO Y, LUO Q, et al. Practical cucumber leaf disease recognition using improved Swin Transformer and small sample size[J]. Computers and Electronics in Agriculture,2022,199:107163.
[12]SHARMA V, TRIPATHI A K, MITTAL H, et al. Weedgan:a novel generative adversarial network for cotton weed identification[J]. The Visual Computer,2023,39(12):6503-6519.
[13]NAZKI H, YOON S, FUENTES A, et al. Unsupervised image translation using adversarial networks for improved plant disease recognition[J]. Computers and Electronics in Agriculture,2020,168:105117.
[14]SINGH A K, RAO A, CHATTOPADHYAY P, et al. Effective plant disease diagnosis using Vision Transformer trained with leafy-generative adversarial network-generated images[J]. Expert Systems with Applications,2024,254:124387.
[15]MAI J Y, LIN H N, HONG X Z, et al. Prediction of potato rot level by using electronic nose based on data augmentation and channel attention conditional convolutional neural networks[J]. Chemosensors,2024,12(12):275.
[16]RADFORD A, METZ L, CHINTALA S. Unsupervised representation learning with deep convolutional generative adversarial networks[EB/OL]. （2015-11-19）
[2025-03-01]. https://doi.org/10.48550/arXiv.1511.06434.
[17]REDDY S R G, VARMA G P S, DAVULURI R L. Resnet-based modified red deer optimization with DLCNN classifier for plant disease identification and classification[J]. Computers and Electrical Engineering,2023,105:108492.
[18]MA R, WANG J, ZHAO W, et al. Identification of maize seed varieties using MobileNetV2 with improved attention mechanism CBAM[J]. Agriculture,2022,13(1):11.
[19]FANG W, GU E M, YI W N, et al. A new method of image restoration technology based on WGAN[J]. Computer Systems Science And Engineering,2022,41(2):689-698.
[20]GULRAJANI I, AHMED F, ARJOVSKY M, et al. Improved training of Wasserstein GANs[C]//NIPS. Proceedings of the 24st international conference on neural information processing systems. Long Beach:Curran Associates Inc.,2017.

备注/Memo

备注/Memo:
收稿日期：2025-03-21基金项目：国家自然科学基金项目(61563046)；塔里木大学研究生科研创新项目(TDGRI202358)作者简介：雷竣杰(2000-),男,山西晋城人,硕士研究生,研究方向为作物信息技术与精准农业。(E-mail)2826100782@qq.com通讯作者：周保平,(E-mail)502805150@qq.com
更新日期/Last Update: 2025-06-24