The diabetic patients may be afflicted by an eye sickness known as diabetic retinopathy (DR) in which excessive blood sugar damages the small blood vessels inside the retina. In the initial stage i.e., Non-proliferative DR (NPDR), fluid starts leaking from blood vessels into retina making it to swell. In this paper, we are presenting an automated technique for detection of four phases: Preprocessing, Feature Extraction, Segmentation and Classification. Pre-processing comprises of removal of noise from the image and converting the image into grayscale. The technique of adaptive histogram can be applied to highlight the hidden portion from the image. In the segmentation process, circular disk segmentation can isolate circular parts from an image. The technique of neural network is applied for classifying diabetic and non-diabetic portion of the image more accurately. The presentation of the proposed work is demonstrated by way of comparing it with existing strategies in phrases of sensitivity, accuracy and specificity.