Semi-supervised assisted cell pathology image multi-target decoupling contrast learning method and system

The invention discloses a semi-supervised assisted multi-target decoupling comparative learning method and system for cell pathology images, and relates to the technical field of computer vision, and the method comprises the steps: introducing a foreground and background decoupling module, splitting features extracted by two enhanced branches into foreground features and background features, and calculating the comparison loss; introducing a hybrid image decoupling branch based on mix, using an isomorphic decoupling module to decouple global features belonging to different instances, and combining with foreground fusion features of two enhanced branches to optimize loss; a semi-supervised training enhancement strategy guided by a small number of semantic tags is provided to assist in initial training and convergence of a decoupling module, and the obtained characterization model is applied to classification detection of cerebrospinal fluid cells. According to the method, the distinguishing capability of an en