An Approach to Implementing Convolutional Neural Network Based on Low Density FPGA

Convolutional neural network (CNN) is an important model in deep learning, which is widely used in image processing. This paper presents a design and implementation of CNN based on low density FPGA by means of SSD and Paddle-Lite architecture. Taking the application of license plate detection as an example, comparing to the traditional target detection method, our experiment shows that the CNN based low cost and low density FPGA can work well in object detection, and it is suitable for some mobile intelligent terminals and embedded systems to perform the task of edge computing.