College of Information and Computer, Taiyuan University of Technology
College of Software, Taiyuan University of Technology
Shanxi Institute of Meteorological Sciences

【目的】为了获得足够的番茄叶片病害图像,提高番茄病害的识别准确率,提出了一种新的基于生成对抗网络的数据增强方法(HiddenparameterlabelandAttentionattachedMultiscaleACGAN,HAM_ACGAN).【方法】在生成对抗网络的基础上,为了补充类内信息,将隐参数标签连接到输入噪声上,以控制不同类别病害的叶片生成;同时设计一个带有残差注意力块的生成器捕获叶片中的病害信息,生成病害特征明显的番茄叶片;最后利用多尺度判别器丰富生成图片的细节纹理。【结论】实验表明,提出的数据增强方法能够生成病害特征明显的番茄叶片,能够满足神经网络训练所需的大量数据,进而提高病害识别网络的识别精度。

【Purposes】 At present, tomato disease recognition based on convolutional neural network relies on a large amount of labeled data, and the lack of data samples is an important problem affecting the accuracy of tomato disease recognition. 【Methods】 Therefore, in order to obtain enough tomato leaf disease images and improve the accuracy of tomato disease recognition, a new data augmentation method HAM_ACGAN (Hidden parameter label and Attention attached Multi scale ACGAN)based on Generative Adversarial Network (GAN)is proposed. On the basis of with auxiliary classifiers, in order to supplement the intra-class information, the hidden varia-ble is connected to the input noise to control the generation of different classes of diseases on the leaves; at the same time, a generator with residual attention block is designed to capture the dis-ease information in the leaves to generate tomato leaves with obvious disease features; finally, a multi-scale discriminator is used to enrich the detail texture of the generated images. 【Conclusions】 The experimental results show that the proposed data enhancement method can generate tomato leaves with obvious disease features, which can meet the large data amount requirement for neural network training, thereby it can improves the recognition accuracy of the disease recog-nition network.

中央引导地方科技发展资金资助项目(YDZJSX2021C005)

罗东升,周子敬,王志伟,等.改进 ACGAN 数据增强的番茄叶片病害识别[J].太原理工大学学报,2023,54(5):861-868.

LUO Dongsheng,ZHOU Zijing,WANG Zhiwei,et al.Tomato leaf disease recognition based on improved AC-GAN data enhancement[J].Journal of Taiyuan University of Technology,2023,54(5):861-868.