Joint Channel Pruning and Quantization-Based CNN Network Learning with Mobile Computing-Based Image Recognition

Huanyu Liu,Jeng-Shyang Pan,Qing Luo,Junbao Li,Mingmei Shao

doi:10.1155/2021/9310558

Abstract

The development of the Internet and communication technology has ushered in a new era of the Internet of Things (IoT). Moreover, with the rapid development of artificial intelligence, objects are endowed with intelligence, such as home automation and smart healthcare, which are typical applications of artificial intelligence technology in IoT. With the rise of convolutional neural network (CNN) in the field of computer vision, more and more practical applications need to deploy CNN on mobile devices. However, due to the large amount of CNN computing operations and the large number of parameters, it is difficult to deploy on ordinary edge devices. The neural network model compression method has become a popular technology to reduce the computational cost and has attracted more and more attention. We specifically design a small target detection network for hardware platforms with limited computing resources, use pruning and quantization methods to compress, and demonstrate in VOC dataset and RSOD dataset on the actual hardware platform. Experiments show that the proposed method can maintain a fairly accurate rate while greatly speeding up the inference speed.

Highlights

In recent years, with the rapid development of the mobile infrastructure of Internet of Things (IoT) and the increasing popularity of the application of IoT, the complexity and operability of various mobile applications have been continuously increasing, and the requirements for the intelligence of mobile applications are getting higher and higher
Artificial intelligence has been gradually applied to all aspects of IoT, such as home automation [1], smart healthcare [2], smart security [3], autopilot [4], and other fields
The popular object detection algorithms are usually based on convolutional neural networks, which are difficult to be deployed on platforms with limited computational resources such as embedded platforms due to the limitation of computational amount and large number of parameters

Summary

Introduction

With the rapid development of the mobile infrastructure of IoT and the increasing popularity of the application of IoT, the complexity and operability of various mobile applications have been continuously increasing, and the requirements for the intelligence of mobile applications are getting higher and higher. It supports structural and unstructured pruning, quantification after and during training, and knowledge distillation methods These functions can only be used in the classification model. The popular object detection algorithms are usually based on convolutional neural networks, which are difficult to be deployed on platforms with limited computational resources such as embedded platforms due to the limitation of computational amount and large number of parameters. In order to be able to deploy the CNN on platforms with limited computing resources, choose a small network designed for mobile devices, and use the compression technique to achieve better results. We use channel pruning and quantization methods to compress the small detection network and test inference time on actual hardware

Proposed Method

Experiments

Findings

Conclusions