In this study, a mobile based system is presented for deep diagnosis of non-proliferative diabetic retinopathy. In this system, firstly, fundus images taken sequentially via the microlens mounted on the smartphone will be sent to the server. The images will then be fused in the server and after some image preprocessing steps, the pathological regions in the fundus image will be classified. For the classification process, the deep convolutional neural network approach will be considered. The system, in its current state, performs detection of hard exudates in the moderate state of non-proliferative diabetic retinopathy in a cloud environment. In the later phases of the study, it is planned to implement a semi-automated mobile-based deep diagnostic system that performs in the cloud environment to detect other abnormalities in abnormal fundus images.