结合注意力机制的相对GAN螺栓图像生成

螺栓缺陷非常容易引起输电线路异常甚至故障,但大量的缺陷数据难以获得。将生成式对抗网络应用于缺陷螺栓图像的生成,针对生成过程中存在的图像质量差、生成样本单一,模型收敛缓慢等问题,提出一种基于改进DCGAN的螺栓图像生成方法。首先在损失函数中加入相对均值鉴别器和梯度惩罚,平衡了生成器和判别器的能力,提高了样本质量和模型的收敛速度；然后在模型的生成器和鉴别器中引入注意力机制,捕获图像中长距离的像素特征,提高了缺陷样本的多样性；实验结果验证了改进方法的有效性,实现了缺陷样本的扩增。

Relativistic GAN for bolts generation with attention mechanism

Bolt defects are very likely to cause abnormal or even faulty transmission lines, but a large amount of defect data is difficult to obtain. This paper applies the generated confrontation network to the generation of defective bolt images. This paper applies the generative adversarial network to the generation of defective bolt images. Regarding the problem of poor image quality, random sample generation, slow model convergence, etc. during the generation, a bolt image generation method based on improved Deep Convolutional Generative Adversarial Networks(DCGAN) is proposed. Firstly, the relativistic average discriminator and gradient penalty are added to the loss function, which balances the ability of the generator and the discriminator, and improves the convergence speed and image quality of the model. Then, the attention mechanism is introduced in the generator and the discriminator, the pixel features of the long distance in the picture are obtained, which improves the diversity of the defect sample. Experimental results verify the effectiveness of the algorithm and achieve the amplification of the defect samples.