摘要
为解决服装数据分析领域中人工标注服装领型效率低的问题,论文提出了一种基于Faster RCNN网络的服装领型识别方法。该方法以Faster RCNN网络框架为基础,ResNet50为主干特征提取网络,SGD优化器与余弦退火算法相结合来优化网络模型,结合迁移学习的方法实现了服装领型的分类和定位。实验结果表明,论文提出方法对于服装领型的识别效果要优于一般的训练方法,map高达95.51%,该方法也为解决服装领域中其他小目标物件识别困难问题提供了新的思路以及解决办法。
In order to solve the problem of low efficiency of manual labeling of garment collar style in the field of garment data analysis,this paper proposes a garment collar style identification method based on Faster RCNN network.The method is based on the Faster RCNN network framework,ResNet50 is the backbone feature extraction network,SGD optimizer and cosine annealing algorithm are combined to optimize the network model,and migration learning is combined to achieve the classification and localization of garment collar styles.The experimental results show that the proposed method is more effective than the general training methods in recognizing the collar type of clothing,with a map as high as 95.51%,and the method also provides new ideas and solutions to solve the problem of difficult recognition of other small target objects in the field of clothing.
作者
李振兴
张迪明
王荦涵
LI Zhenxing;ZHANG Diming;WANG Luohan(Jiangsu University of Science and Technology,Zhenjiang 212100)
出处
《计算机与数字工程》
2025年第6期1752-1756,共5页
Computer & Digital Engineering
作者简介
李振兴,男,硕士研究生,研究方向:计算机视觉;张迪明,男,讲师,研究方向:人工智能、高性能计算;王荦涵,男,硕士研究生,研究方向:计算机视觉。
