改进DCGAN的飞机蒙皮图像生成方法

Aircraft skin image generation method of improved DCGAN

针对传统深度卷积生成对抗网络(DCGAN)在生成飞机蒙皮图像中存在图像质量差和训练不稳定的问题,提出了一种改进的生成器网络。利用ResNet残差模块改进了DCGAN生成器与判别器结构,以解决因网络加深和图像尺寸增大导致的生成图像质量差的问题;采用Wasserstein距离作为新的损失函数,以增强网络的训练稳定性。试验表明,改进后的模型训练稳定性得到增强,生成的飞机蒙皮图像的SMD值提升了30.6%,Tenengrad梯度值提升了41.5%,Laplacian梯度值提升了13.4%。

Aiming at the problems of poor image quality and unstable training in the generation of aircraft skin images by traditional deep convolution generated adversarial network(DCGAN),an improved generator network was proposed.The structure of DCGAN generator and discriminator was improved by using ResNet module to solve the problem of poor image quality caused by deepening network and increasing image size.Wasserstein distance was used as a new loss function to enhance the training stability of the network.Experimental results show that the training stability of the improved model is enhanced,and the SMD value of the generated aircraft skin image is improved by 30.6%,Tenengrad value is improved by 41.5%,and Laplacian value is improved by 13.4%.