Accurate detection of road surface pothole distress is crucial for road safety operation and intelligent maintenance.To address the issues of scarce pothole samples

imbalanced class distribution

and insufficient resolution of existing generative models in real-world scenarios

a data augmentation method for road surface distress using a vehicle-mounted camera is proposed

combining residual deep generative adversarial network(GAN)with hybrid gradient Poisson fusion.A residual Wasserstein in GAN with gradient penalty(ResWPGANet)generative model is constructed

incorporating deep residual structures and progressive upsampling to enhance the capability of generating high-resolution pothole images.A gradient penalty mechanism is adopted to ensure training stability.Hybrid gradient Poisson fusion is used to seamlessly embed the generated pothole regions into real road surface backgrounds to enhance sample realism.Experimental results based on the RDD2022 dataset demonstrate that the proposed method outperforms mainstream generative models. After the generated samples are used to train the detection model

the pothole category detection accuracy improves by 4.77%

and the overall detection accuracy improves by 1.42%.This method effectively alleviates the sample imbalance problem and provides technical support for intelligent vehicle-mounted road surface distress detection.