Normalization algorithms and implementations used in deep learning

The most effective use of the data set has a critical structure in order for artificial intelligence applications to exhibit high performance. Normalization annotations are techniques that aim to eliminate outliers and prevent the network from displaying high bias by arranging datasets for various countries. Apart from this, in addition to the existing normalization measures, it is aimed to perform the COVID-19 inspection with the scanned radiography image dataset and the non-covered CNN model by using MVSR Normalization, which has been suggested as a new method in the literature, and to increase the performance of the model of this method. Eliminating this situation by using the COVID-19 Radiography Database, the CNN model in the Google Colab environment was fed with scan radiography image datasets without normalization, Min-Max, MVSR, MVSR+Min-Max Normalizations applied, respectively, the models were to be implemented in the Kria KV260 Vision AI Starter Kit environment and Each of the models was evaluated with performance metrics that were tested in the kit environment. MVSR Normalization was performed in Kria KV260 Vision AI Starter Kit environment, and Min-Max Normalization was performed in Colab environment. At the end of the study, the model with the highest success was tested again using the kit's camera. In this thesis, the highest performance was obtained with MVSR+Min-Max Normalization as 95%.