Automotive condenser image generation method based on improved DCGAN

Deep learning methods have made progress in the field of industrial product image defect detection, but it is difficult to collect a large number of defect data. The existing methods have problems on generating automotive condenser defect images, such as low image generation quality, inability to generate images according to defect categories, and slow model convergence speed. Applying generative adversarial network to defect image generation, adeep convolution generative adversarial network (DCGAN) model based on semi-supervised and self-attention mechanism is proposed togenerateautomotivecondenserdefectimages. Self-Attention mechanism is introduced in DCGAN to overcome the problem of long-range feature extraction in convolutional network and improve the quality of generated samples. By semi-supervised learning, a supervised classifier is added into the unsupervised discriminator, the cross entropy loss of the classifier and gradient penalty are added into the loss function of the discriminator, which improves the convergence speed and stability of the model. Condition normalization is used to adjust convolution layer parameters, and the defect category information of images is embedded into the discriminator, which improves the diversity of generated samples and enable the model to generate condenser images with specific defects. Experimental results show that the proposed method can generate high-quality automotive condenser defect images, the FID value reaches 43.7, which is better than the existing DCGAN and SAGAN. Compared with ACGAN, the diversity of generated images is also significantly improved.