The 15th IEEE International Conference on Electron Devices and Solid-State Circuits (EDSSC)

624 / 2019-03-15 23:59:35

Binarized Convolutional Neural Networks Converted to SNN for Object Detection

CNN,binary neural networks,object detection,snn

Plenary track > Algorithm

In the real world, object detection algorithms sometimes need to be deployed to embedded mobile terminals, and low-latency, low-power real-time detection is needed. At the same time, using tools such as Tensorflow, it is proved by experiments that the binarized neural network can accelerate and compress the network model without greatly reducing the performance The correlation model and training algorithm of Spiking depth neural network are introduced. An algorithm model of convolutional neural network Spiking is proposed. A Spikingd neural network model based on binarization is realized by using tools such as Tensorflow and SNN Toolbox. The experimental results are verified and analyzed, and finally provide a new idea for the Spiking neural network hardware circuit configuration.