The invention relates to an on-orbit detection method for a steel rail in a running state. The detection method comprises the steps of information acquisition, image acquisition, image preprocessing, judgment and classification of a neural network model, regression analysis of a multiple linear regression model, relative error calculation and information transmission. Real-time detection of the damage condition of the steel rail in the running state is achieved, interpretation is conducted through the neural network algorithm with the deep learning capacity, the confidence coefficient and stability of interpretation are improved, regression analysis is conducted on the interpretation result through the multiple linear regression model, the influence of train running parameters and meteorological parameters on the interpretation result is effectively avoided, and the detection result is closer to the real situation; and meanwhile, the invention further provides detection equipment based on the steel rail on-orbit detection method, and compatible sharing, intercommunication and interconnection with a train control system are effectively achieved through equipment carrying, wire harness integration and program implantation.