基于深度残差收缩网络的商品图像识别  

Commodity Image Recognition Based on Deep Residual Shrinkage Network

作　　者：李昊璇[1] 闫新艳 LI Haoxuan;YAN Xinyan(College of Physics and Electronic Engineering, Shanxi University, Taiyuan 030006, China)

机构地区：[1]山西大学物理电子工程学院,山西太原030006

出　　处：《测试技术学报》2021年第4期294-299,322,共7页Journal of Test and Measurement Technology

摘　　要：为了降低噪声信息的干扰及提高商品图像识别的准确率,提出了基于深度残差收缩网络的商品图像识别模型.该模型在深度残差网络的基础上融入软阈值函数及注意力机制,软阈值函数将注意力机制注意到的不重要的特征置为0,从而降低噪声信息的干扰,提高图像识别的准确率.实验首先通过爬虫方式获取了包含了51种商品的数据集,并且对该数据集通过图像翻转以及对图像加噪等操作,形成具有44066张图像的商品数据库.然后将深度残差收缩网络与深度残差网络、SENet算法模型对数据进行训练对比,同时对部分商品图像进行了测试.实验结果表明,深度残差收缩网络不仅可以提高商品图像识别准确率,同时还提高了模型的运行速度.In order to improve the accuracy of image recognition and reduce the interference of noise information,this paper proposed a commodity image recongnition model based on deep residual shrinkage network.The soft threshold function and attention mechanism were integrated into the model based on deep residual network.The soft threshold function set the unimportant features noticed by attention mechanism to 0,so as to reduce the interference of noise information and improve the accuracy of image recognition.Initially,the data set containing 51 kinds of commodities was obtained by crawler,and the commodity database with 44066 images was formed by image flipping and image image with noise.Then,the deep residual shrinkage network was compared with the deep residual network and the SENet algorithm model,and some commodity images are tested.Experimental results showed that the deep residual shrinkage network can improve the accuracy of commodity image recognition and the running speed of the model.

关 键 词：商品识别 深度残差收缩网络 注意力机制 软阈值函数 噪声 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]