摘要
单发多框检测器算法(Single Shot Multibox Detector,SSD)采用多个特征层进行目标检测,但每一层都是独立使用的,这种结构忽略了上下文信息,不利于提高小目标检测的精度。为了提高传统SSD算法精度,提出了一种特征信息增强的SSD算法(Feature Information Enhancement Based Single Shot Multibox Detector,FESSD),其核心是一个特征信息增强模块。首先提出一个特征融合模块来对不同特征层进行融合和细化。然后采用一种挤压和激励模块(Squeeze and Excitation block)来自适应地获取每个特征通道的重要程度,从而增强有用信息和抑制无用信息。最后仿真结果表明,相比于传统SSD算法,FESSD算法能够有效地提升目标检测的精度。
Single Shot Multibox Detector(SSD)takes several feature layers for object detection,but each layer is used independently.This structure may ignore context information and is not conducive to optimize the detection accuracy of small objects.In order to improve the performance of SSD,a Feature Information Enhancement Based Single Shot Multibox Detector(FESSD)is proposed,and the core is a feature enhanced module.Firstly,a feature fusion module is proposed to fuse and refine the different feature layers.Then,a squeeze and excitation block is adopted to acquire the importance of each channel adaptively,so as to enhance the useful information and suppress the useless information.Finally,the simulation results show that,compared with the conventional SSD algorithm,the FESSD algorithm can improve the accuracy of object detection effectively.
作者
赵辉
李志伟
方禄发
ZHAO Hui;LI Zhiwei;FANG Lufa(School of Communication and Information Engineering,Chongqing Universityof Posts and Telecommunications,Chongqing 400065,China;Chongqing Key Laboratory of Signal and Information Processing,Chongqing University of Posts and Telecommunications,Chongqing 400065,China)
出处
《计算机工程与应用》
CSCD
北大核心
2021年第4期148-154,共7页
Computer Engineering and Applications
基金
国家自然科学基金(61671095)。
关键词
目标检测
特征融合
注意力机制
深度学习
object detection
feature fusion
attention mechanism
deep learning