A composite insulator exhibits different heating characteristics under different defect types. In this study, a thermal defect detection method for composite insulators based on one-dimensional residual network and the central axis temperature data of composite insulators is proposed. First, the abnormal temperature rise range and position information of composite insulators under different defect types were statistically analyzed, and a sample set of composite insulator central axis temperature data under different defect types was obtained. Next, a one-dimensional residual network model was established. The dilated convolution was introduced into the residual block to expand the receptive field, and the efficient channel attention network (ECA_Net) was added to improve the feature weight with a high correlation with the defect category. Finally, numerical examples were verified and compared. Simultaneously, the t-distributed stochastic neighbor embedding (t-SNE) visualization method was used to reflect the effect of feature extraction on the model. The results showed that the model effectively captured the spatial dimension information of the central axis temperature data and adaptively extracted the features with high classification discrimination. Compared with ordinary convolution, auto encoder (AE), and support vector machine(SVM), the proposed model has improved recognition accuracy, good robustness and generalization ability. Thus, end-to-end composite insulator heating defect detection was realized.