The present invention provides a bridge structure health diagnosis method based on deep learning. The method comprises: step one, preprocessing data; step two, by using data in a design stage and a construction stage, establishing a precise finite element model of a bridge structure as a training input of a deep learning network; step three, constructing a five-layer deep learning network DN; step four, fine-tuning a fifth layer of the deep learning network; step five, input another part of data in a monitoring stage into the network to further optimize the whole deep learning network; and step six, according to the deep learning network that is formed in step five and can be used for structure health diagnosis, diagnosing whether a structure health state of any one bridge is normal or in which deterioration state level. According to the bridge structure health diagnosis method based on deep learning provided by the present invention, the health state of various bridge structures can be diagnosed and checked quickly and effectively by establishing the deep learning network.