Hot rolled steel strip surface defect detection method based on semi-supervised transfer learning

The invention discloses a semi-supervised transfer learning-based hot rolled steel strip surface defect detection method, which comprises the following steps of: 1, adjusting, checking and cleaning mark information in an existing hot rolled steel strip data set NEU-DET by using LabelImg mark software, and dividing six types of defect images in the data set NEU-DET into a training set, a verification set and a test set; 2, performing data enhancement on the training set to realize data expansion; 3, constructing and training a target detection model; 4, taking the improved Faster R-CNN algorithm as a detection model, introducing Mosaic data enhancement, a KeepAugent method and an EMA method on the basis of strong data enhancement of a semi-supervised learning framework STAC to obtain an improved semi-supervised learning framework STAC, performing strong data enhancement and weak data enhancement on unmarked hot rolled steel strip data, and inputting weak data enhancement into a teacher model, so as to obtain a