Aero-engine blade coding dot matrix character recognition method based on deep transfer learning

The invention relates to an aero-engine blade coding dot matrix character recognition method based on deep transfer learning, and belongs to the technical field of industrial character recognition. A blade image is obtained through an industrial camera, positioning extraction of a blade tenon coding dot matrix character region is completed by using an existing universal target detection network model, and image enhancement and augmentation are performed on a coding dot matrix character region image by means of multiple image processing methods to complete construction of a target domain data set. An automobile part picture data set with industrial character identification codes of a homologous task and a self-generated clear imitation code data set are used for processing to complete source domain data set construction, a YOLO-GhostNetV2-EMA network model structure diagram is designed and constructed, and a deep transfer learning fine adjustment experiment strategy is designed according to the provided model.