Deep neural network used for recognizing diabetes retinopathy and system

The invention provides a deep neural network used for recognizing diabetes retinopathy. The deep neural network comprises a preprocessing module used for respectively preprocessing a target fundus image and a reference fundus image from the same person, wherein the target fundus image and the reference fundus image belong to different eyes; a first neural network used for generating a first advanced feature set from the target fundus image; a second neural network used for generating a second advanced feature set from the reference fundus image; a feature combination module used for combing the first advanced feature set and the second advanced feature set to form a feature combination set; and a third neural network used for generating a pathological determination result according to the feature combination set. According to the deep neural network, the target fundus image and the reference fundus image which belong to different eyes are respectively and independently employed as input information so that the determination accuracy of the pathology of the fundus images can be improved.