The invention discloses a
deep learning-based early age-related macular
lesion weakly supervised classification method, which comprises the following steps of 1, positioning a central concave positionof an eye
fundus image by adopting a
convolutional neural network, and intercepting a square area as a candidate area by taking the central concave position as an original point and taking a double-
optic disc diameter as a side length; 2, judging whether glass membrane warts appear in the macular area or not by adopting a
convolutional neural network, detecting the glass membrane warts in a weaksupervision manner, and judging whether the glass membrane warts appear in the
fundus image or not; 3, performing linear interpolation by using the intermediate result of the step 2 to obtain a finalpixel-level focus marking result. According to the
algorithm, a weak supervision method is adopted for classifier training and detection, only whether the
fundus image has vitreous condyloma information or not needs to be provided, the classifier can be trained without specific position information, correct classification of the early-stage age-related macular
lesion fundus image is achieved, andthe
algorithm can effectively save the cost of marking training data while the precision is guaranteed.