Abstract: The defect detection for the metal casing of solid rocket motor is essential in the assessment of motor conditions. A machine vision-based method for detecting surface defects on metal casings of solid rocket motors was proposed in this paper. The method employed data augmentation using deep convolutional generative adversarial networks (DCGAN), and target detection technique using a convolutional block attention module (CBAM). By utilizing limited real defect data and deep learning algorithms to augment the dataset, the method identified and classified defective targets. The method was tested against various types of casing defects, showing a 5.7% improvement in identification accuracy over conventional detection methods, while maintaining detection efficiency. The model also exhibits good robustness and generalization, indicating potential for practical engineering applications.