基于VGGNet-16的满族服饰识别研究  

Recognition of Manchu Clothing Based on VGGNet-16

作　　者：金丹 邹珍 JIN Dan;ZOU Zhen(Shenyang Light Industry Art School,Shenyang 110031,China;Hangzhou Hongshengqi Technology Co.,Ltd.,Hangzhou 311199,China)

机构地区：[1]沈阳市轻工艺术学校服装系,沈阳110031 [2]杭州鸿盛启科技有限公司,杭州311199

出　　处：《纺织科技进展》2024年第2期23-28,43,共7页Progress in Textile Science & Technology

基　　金：辽宁省职业技术教育学会项目(LZYZXJSYB2109)。

摘　　要：针对满族服饰在识别过程中,特征分类不准确且识别精确度低的问题,以满族旗袍为特征识别标准,优化设计VGGNet-16与特征处理,完成对满族旗袍的识别。首先,在VGGNet-16框架中,通过图像预处理、模型构建与特征提取,完成对满族旗袍的粗略识别;然后,优化特征处理,筛选纹理特征,完成对训练集的分类与预测;最后,实现准确的服饰识别。通过与基于阈值分割的服饰识别算法、基于改进的DenseNet-BC服饰识别算法对比得知,识别精确度分别提高6.35%、7.18%,分类精确度分别提高6.71%、6.50%。In response to the problem of inaccurate feature classification and low recognition accuracy in the recognition process of Manchu clothing,the Manchu cheongsam was taken as the feature recognition standard,the design of VGGNet-16 and feature processing were optimized,while the recognition of Manchu cheongsam was completed.Firstly,in the VGGNet-16 framework,rough recognition of Manchu cheongsam was achieved through image preprocessing,model construction and feature extraction.Afterwards,feature processing was optimized,texture features were filtered,while the classification and prediction of the training set were completed.The accurate clothing recognition was achieved.By comparing the algorithm with the clothing recognition algorithm based on threshold segmentation and the improved DenseNet-BC clothing recognition algorithm,it was found that the recognition accuracy had been improved by 6.35%and 7.18%,while the classification accuracy had been improved by 6.71%and 6.50%.

关 键 词：满族服饰 卷积神经网络 特征提取 

分 类 号：TS941.2[轻工技术与工程—服装设计与工程]