基于空间和ECA注意力机制的废旧手机型号识别研究  

Research on Model Recognition of Used Mobile Phones Based on Spatial and ECA Attention Mechanism

作　　者：余光海 付勇刚 邓梅玲 

机构地区：[1]合肥工业大学机械工业绿色设计与制造重点实验室,合肥230009 [2]中国电器科学研究院股份有限公司,广州510300

出　　处：《日用电器》2022年第11期71-75,共5页ELECTRICAL APPLIANCES

基　　金：国家重点研发计划“固废资源化”专项-废旧移动终端无损检测与评估分类技术及装备(2018YFC1902301)资助。

摘　　要：针对废旧手机的分类回收难以快速精准识别的问题,本文基于残差网络模型Resnet 34进行改进,让模型能自适应地关注到感兴趣的特征和定位到具有判别性的区域,从而提高手机型号识别的准确率。首先,在Resnet 34模型的Layer 4上添加ECA注意力机制,并实验将ECA通道注意力机制与空间注意力机制结合,组成新的Eca-S-Resnet 34注意力机制。然后对网络模型进行训练,基于验证集测试网络模型预测的准确率,最后导入照片测试手机型号识别速度与准确率。实验证明在模型的Layer 4上添加的ECA注意力机制与Eca-S-Resnet 34注意力机制均可以提高网络训练速度与模型的分类准确率。Aiming at the problem that the classification and recycling of used mobile phones are difficult to identify quickly and accurately,this paper improves on the residual network model Resnet 34,so that the model can adaptively focus on the features of interest and locate the discriminative areas,thus improving the accuracy of mobile phone model identification.First,ECA attention mechanism was added to layer4 of Resnet 34 model,and ECA channel attention mechanism was combined with spatial attention mechanism to form a new Eca-S-Resnet 34 attention mechanism.Then the network model is trained,and the accuracy of network model prediction is tested based on the verification set.Finally,photos are imported to test the speed and accuracy of mobile phone model recognition.Experiments show that the ECA attention mechanism and Eca-S-Resnet 34 attention mechanism added to layer 4 of the model can improve the network training speed and the classification accuracy of the model.

关 键 词：手机型号识别 残差网络结构 注意力机制 废旧手机回收 

分 类 号：X705[环境科学与工程—环境工程] TN929.53[电子电信—通信与信息系统]