With the development and progress of artificial intelligence technology, all aspects of our daily life have undergone tremendous changes, especially the application of convolutional neural networks in the fault detection of power conductors in power transmission lines, which has greatly eliminated power transmission. Potential safety hazards ensure people’s electricity consumption, but the application of conventional convolutional neural networks to visual tasks requires a lot of training data, and certain defects in the transmission line are extremely scarce. Collecting and marking these training data consumes huge manpower and material resources. Based on this, this paper proposes to apply self-supervised representation learning algorithm to the fault classification and recognition task of power conductors in transmission lines to alleviate the problem of data labeling difficulties. The self-supervised representation learning algorithm can learn from unlabeled samples and does not require negative sampling. It has higher training efficiency. In the experiment, the self-supervised representation learning algorithm is compared with other baseline methods and its performance is excellent. In the task of classification and identification of wire damage, the average accuracy can reach 0.87, which shows the effectiveness and practicability of the algorithm.