The split pins of highspeed railway catenary positioning tube are easy to be loosed in the longterm running vibration of the train. However, the number of loose samples is scarce. To solve these problems, this study proposes a threelevel cascade architecture expand the defect samples based on deep convolutional generative adversarial network (DCGAN). Then, the convolutional neural network (CNN) is trained to detect split pins defect. Firstly, according to the central point method, the same size image of split pins for training is extracted. Then, DCGAN is used to generate simulated defect samples and a lightweight CNN network is formulated to screen the generated samples. Finally, the extended defect sample set and the positive sample set are utilized to train the detection model on the adjusted VGG16 convolutional neural network. In this way, the defective pins defect state detection can be realized. Experimental results show that the proposed method can achieve 99% accuracy in split pin defect detection of catenary positioning tube.