基于孪生卷积神经网络的手机型号识别方法  

Mobile Phone Model Recognition Method Based on Siamese Convolutional Neural Network

作　　者：韩红桂 甄琪 任柯燕 伍小龙 杜永萍 乔俊飞 HAN Honggui;ZHEN Qi;REN Keyan;WU Xiaolong;DU Yongping;QIAO Junfei(Beijing University of Technology,Faculty of Information Technology,Beijing Key Laboratory of Computational Intelligence and Intelligent System,Beijing,100124,China)

机构地区：[1]北京工业大学信息学部,计算智能与智能系统北京市重点实验室,北京100124

出　　处：《北京工业大学学报》2021年第2期112-119,共8页Journal of Beijing University of Technology

基　　金：国家重点研发计划资助项目(2018YFC1900800-05);国家自然科学基金资助项目(61890930-5).

摘　　要：针对废旧手机回收过程中型号难以精确识别的问题,提出一种基于孪生卷积神经网络的废旧手机型号识别方法.首先,利用基于最大类间差分的边缘检测算法解析手机图像的区域特征,构建手机型号识别数据库;其次,构造一种共享权值孪生卷积网络(siamese convolutional neural network,S-CNN)的手机识别模型,实现废旧手机图像特征的快速提取;最后,设计一种自适应学习率的识别模型参数更新策略,提高手机型号识别的精度.将其应用于不同场景下废旧手机的分拣,实验结果表明该方法具有较好的快速性和准确性.Mobile collection of waste electrical and electronic equipment(WEEE)is convenient for residents and companies.Aiming at the problem that the model is difficult to be accurately identified during the recycling process of used mobile phones,a method for identifying the model of used mobile phone was proposed in this paper based on convolutional neural network.First,the edge detection algorithm was used to construct the mobile phone model of feature database of identification area.Second,a mobile phone recognition model of siamese convolutional neural network(S-CNN)was constructed to obtain the fast feature of used mobile phone images.Finally,an adaptive learning rate mobile phone recognition model parameter training method was designed to improve the accuracy of mobile phone model recognition.The method was applied to the sorting of used mobile phones in different scenarios.The experimental results show that the method has good speed and accuracy.

关 键 词：孪生卷积神经网络 相似性评估 模型参数更新 手机型号识别 边缘检测 废旧手机回收 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TP183[自动化与计算机技术—计算机科学与技术]