The segmentation step of retinal blood vessel helps to diagnosis the diseases including diabetic retinopathy, glaucoma, etc. The automatic image segmentation process helps experts to speed up the diagnosis of DR, since analytic methods are time consuming and error prone. The neural network (NN) based methods like U-Net uses leap bonding that extract fine information from the training dataset. However automatic segmentation of image using neural network is a challenging process because of uneven and irregular geometry of organ. In this article, we proposed a U-Net based approach for segmentation of retinal vessels. Before applying segmentation step, the affected area of image is enhanced with some preprocessing techniques. Then a dual tree discrete Ridgelet transform (DT-DRT) is apply on the dataset to extract the features from the region of interest. The features accumulation with DT-DRT ensures better feature representation of vessel for segmentation task. The proposed segmentation is implemented on different publicly available dataset and achieve accuracy of 96.01% in CHASE DB1, 97.65% in DRIVE and 98.61% in STARE dataset. The performance of this algorithm is also compared with some other deep learning models, and results demonstrate that this proposed algorithm performed better than them.