THE EFFECT OF ADDING ARTIFICIAL NOISE ON THE SUCCESS OF A DEEP LEARNING NETWORK

Abstract

Image pre-processing methods have a significant impact on increasing the success of the deep learning network. For deep learning networks to be successful, a large amount of data is needed. Obtaining these datasets is a challenging task that requires effort and observation. Therefore, the researchers aim to increase the success of the deep learning network by modifying the existing images and making it more prepared for new test images. In this way, they artificially increase the number of images in the data sets by using methods such as modifications to the colour bands, adding noise, and histogram equalization. In this study, the change in success rates of the network will be investigated in detail by adding different noise types to the data set used to train the deep learning network.