New coronal pneumonia CT image classification method based on semi-supervised transfer learning

The invention discloses a new coronal pneumonia CT image classification method based on semi-supervised transfer learning. The method mainly comprises the three steps that focus information is segmented through an in-net network; pixel-level fusion of the focus and the original image is carried out; and the semi-supervised transfer learning framework classifies the fused image. According to the method, firstly, focus segmentation is carried out on a CT data set of a Hongshan hospital by using an infnet network, meanwhile, image fusion is carried out, focus representation of an original image is highlighted, and then a semi-supervised transfer learning framework is trained and used for classifying diseases of a new coronal pneumonia CT image into common diseases, severe diseases and critical diseases. According to the method, segmentation serves as an auxiliary task, picture information is fully utilized, external marking information is not needed, an original CT data set which is completely not marked is clas