Image recognition using capsule network on FPGA

Abstract

A capsule neural network (CapsNet) is a new approach in artificial neural network (ANN) that produces a better model hierarchical relationship. A capsule is a set of neurons. Each capsule generates vector which presents the details of an entity. The performance of CapsNet on graphics processing unit (GPU) is considerably better than convolutional neural network (CNN) at recognizing highly overlapping digits in images. Nevertheless, this new method has not been designed as accelerator on field-programmable gate array (FPGA) to measure the speedup performance and compared it with the GPU. This is because of the lack of hardware design experience. This project aims to design the CapsNet model (accelerator) on FPGA using high-level synthesis (HLS). Then, the performance between FPGA and GPU will be compared, mainly in terms speedup and accuracy. Behavioural module is synthesized using HLS tools on FPGA then it is evaluated and validated using MNIST dataset. The module is designed to receive features vectors of handwritten digits image as an input and pass it through several layers to predict the output. The speed-up performance on FPGA is expected to be higher than GPU, but FPGA accuracy is expected to be slightly lower than GPU. The module can be useful in detecting the license plate of fast-moving vehicles and many other applications.