期刊论文

Image classification algorithm of plant diseases and pests based on semi-supervised generative adversarial networks

中文

武汉轻工大学学报

2095-7386

2019

38

4

45-52

10.3969/j.issn.2095-7386.2019.04.010

本校为第一机构

提出了一种应用于植物病虫害图像识别的模型。利用训练后的半监督生成对抗网络,只需输入随机噪声,即可自动产生植物病虫害的仿真图像。为了验证模型的图像生成和识别能力,分别在MNIST数据集和Plant Village数据集上进行实验。实验表明,通过100次迭代训练,输入噪声向量到生成网络,可以得到高真实度的样本图像,使用判别网络提取特征并进行图像分类实验,在测试集上得到的正确率分别为98.96%和94.32%,取得了较好的分类效果。

In this paper, a model applied to the image classification of plant diseases and insect pests is proposed. Using the trained semi-supervised generative adversarial networks, only the random noise as input of generative network is demanded for the automatic generation of the image of plant diseases and insect pests. For the sake of verifying the image generation and recognition ability of the model, respectively, the experiments were carried out on the handwritten digital dataset(MNIST) and the Plant Village dataset. Experiments shows that after 100 iterations of training, high-fidelity sample ...