The identification of eye disorders using fundus images poses a substantial challenge to the medical field. In fact, distinct phases of severity are shown by each pathology, and these stages may be deduced by confirming presence of lesions and defining their morphological characteristics. Moreover, several lesions with distinct diseases share traits with one another. Several techniques have been put out for using fundus images to identify eye conditions. Because deep learning (DL) based techniques may tailor the network to the desired detection outcome, they outperform other methods in terms of detection. This paper identifies ocular diseases using DL based techniques. Adaptive Wiener filter is first used to resize, crop, reflect, and eliminate noise from the image. Subsequently, SMOTE is used to address data imbalances and perform data augmentations. Lastly, a method for illness identification called Federated Deep Learning (FDL) is suggested. We used FDL to four pre-trained models, including convolutional neural network (CNN), VGG16, VGG19, and ResNetV2. DL models taught centrally are contrasted with those learnt within the federated framework.