Infant retinal disease information identification method and system based on training network

The invention relates to an infant retinal disease information identification method and system based on a training network, and the method comprises the following steps: employing a CNN and Transform hybrid network to extract local-global features of a to-be-detected retinal image, employing a deep attention fusion module to carry out the fusion of the local-global features, training ROP data in the fused features by adopting a training network, extracting information with deep feature expression, and realizing ROP severity grading; the advantages and disadvantages of the CNN and the Transform network are combined to realize double tasks of automatically detecting various infant retinopathy and grading ROP severity; various common infant fundus lesions can be detected, the influence of other lesion characteristics is shielded, and the accuracy of the automatic examination system for the common infant fundus lesions is greatly improved; the discomfort caused by difficult diagnosis and repeated examination of