轻量化卷积神经网络目标检测算法综述

A Survey of Object Detection Algorithms for Lightweight Convolutional Neural Networks

随着近年来深度学习的迅猛发展,基于卷积神经网络(CNN)的目标检测算法由于相比传统算法更具优势,已成为当前解决目标检测问题的首选方法。为了实现CNN类目标检测算法在嵌入式等资源受限的平台上高效运行,各种轻量化CNN类网络和目标检测算法近年来得到了快速发展。首先对SqueezeNet、MobileNet、ShuffleNet三类轻量化网络进行了概述,在此基础上对CNN类轻量化目标检测算法发展历程和研究进展进行了系统总结,并结合各类算法在VOC等数据集上的测试结果对其进行了对比分析,讨论了各类轻量化目标检测算法的性能特点及未来发展方向。

With the rapid development of deep learning in recent years,the object detection algorithm based on convolutional neural network(CNN)has become the preferred method to solve the object detection problems because it has significant advantages over traditional algorithms. In order to realize the CNN class object detection algorithms efficient implementation on embedded platforms such as resource constraints,various lightweight CNN class networks and object detection algorithms have developed rapidly recently. Firstly,the typical object detection algorithms of R-CNN、YOLO and SSD are summarized based on candidate regions and regression method. On this basis,SqueezeNet,MobileNet and ShuffleNet three types of lightweight networks are summarized,and the development history and research progress of CNN class lightweight object detection algorithms are systematically summarized in combination with the test results of various algorithms on VOC and other sets of data,it is compared and analyzed. The performance characteristics of lightweight object detection algorithms and the future development direction are discussed.