This paper discusses the mathematics behind convolution and Fourier transformations. Experiments are presented that support the integration of a two-dimensional discrete fast Fourier transformation layer into a convolutional neural network to minimize the computational costs of solving computer vision problems.