摘要
针对垃圾分类检测任务中检测目标尺寸不一和小目标检测精度不高等问题,构建一种基于隐式特征金字塔网络(IFPN)和MobileNetV2的改进SSD模型的分类检测方法,对垃圾进行实时分类检测。首先,将改进后的MobileNetV2引入SSD,加入带有空洞卷积的空间金字塔池化模块(ASPP),在降低网络模型计算复杂度的同时保证网络实时性和精确性;其次,采用IFPN从网络的深层到浅层逐级融合SSD,更精确地检测出小目标;最后,使用Focal Loss函数调节正负样本之间的权重。实验结果表明,在阈值为0.4时,所提方法比传统SSD平均精确率均值(mAP)提高了4.84个百分点,检测耗时减少了72.7%,能满足边缘计算设备对模型的各项要求。
Aiming at the problems of different target sizes and low detection accuracy of small targets in the task of garbage classification and detection,a classification and detection method based on the improved SSD(Single Shot multiBox Detector)model of Implicit Feature Pyramid Network(IFPN)and MobileNetV2 was constructed to carry out real-time classification and detection of garbage.Firstly,the improved MobileNetV2 network was introduced into SSD by adding an Atrous Spatial Pyramid Pooling module(ASPP)to ensure real-time network performance and accuracy while reducing the computational complexity of the network model.Secondly,IFPN was used to fuse SSD from deep to shallow layers of the network step by step to detect small target more accurately.Finally,Focal Loss function was used to adjust the weights between positive and negative samples.The experimental results show that at a threshold of 0.4,the proposed method improves the average accuracy rate by 4.84 percentage points and reduces the detection time by 72.7%compared to the traditional SSD,meeting all the requirements of the model for edge computing devices.
作者
赵珊
刘子路
郑爱玲
高雨
ZHAO Shan;LIU Zilu;ZHENG Ailing;GAO Yu(School of Computer Science and Technology,Henan Polytechnic University,Jiaozuo Henan 454003,China)
出处
《计算机应用》
CSCD
北大核心
2022年第S01期106-111,共6页
journal of Computer Applications
基金
国家自然科学基金资助项目(61572173)
河南省高等学校重点科研项目(18B520017)
河南理工大学博士基金资助项目(B2014⁃043)。
关键词
垃圾分类
目标检测
MobileNetV2
SSD
空间金字塔池化
隐式特征金字塔网络
garbage classification
target detection
MobileNetV2
Single Shot MultiBox Detector(SSD)
Spatial Pyramid Pooling(SPP)
Implicit Feature Pyramid Network(IFPN)
