An Efficient CNN Architecture for Image Classification on FPGA Accelerator

Image classification finds its suitability in applications ranging from medical diagnostics to autonomous vehicles. The existing architectures are computationally exhaustive, complex and less accurate. An accurate, simple and hardware efficient architecture is required to be developed for image classification. In this paper, Convolutional Neural Network (CNN) architecture has been proposed and validated using MNIST handwritten dataset. The adopted approaches of sliding-filter for convolution and parallel computation of Multiplication and Accumulation (MAC) operations resulted in optimized hardware architecture with reduced arithmetic operations and faster computations. The developed architecture has been implemented on Artix-7 FPGA and attained a significant improvement in speed compared to existing architecture working at 300MHz maximum operating frequency.