Conveyor belt damage detection model training method, application method and medium

The invention discloses a conveyor belt damage detection model training method, an application method and a medium, and the method comprises the steps: obtaining a conveyor belt image training sample set containing annotation information, inputting the conveyor belt image training sample set to a constructed conveyor belt surface damage detection network, determining a corresponding conveyor belt image prediction sample set, and carrying out the detection of the conveyor belt surface damage. Determining a loss function according to the conveyor belt image training sample set and the conveyor belt image prediction sample set; the conveyor belt surface damage detection network comprises an improved YOLOX network; and according to the value of the loss function, adopting a preset cross-domain migration and intra-domain migration method to adjust parameters of the conveyor belt surface damage detection network until convergence conditions are met, and determining the completely trained conveyor belt surface damag