Dynamic configuration of FPGA-based convolutional neural network parameters

Convolutional neural networks (CNNs) have been widely used in the field of target recognition, and it has become a challenge to deploy convolutional neural networks with high accuracy and the ability to recognize multiple targets on a single chip. In order to save FPGA resources and improve target recognition accuracy with limited resources, this paper proposes an idea of constantly changing the weights (w) and biases (b) of the convolutional and fully connected layers in order to recognize different kinds of targets, and dynamically configuring w and b into the neural network to achieve the effect of improving target recognition accuracy[1]. The improved network is deployed on a Xilinx ZYNQ xc7z020 FPGA, which effectively improves the target recognition accuracy with fast recognition speed and low power consumption.