东华大学学报（英文版）2025，Vol.42Issue(3)：301-314,14.DOI:10.19884/j.1672-5220.202411017
基于深度学习的时尚领型实时检测方法
A Real-Time Detection Method for Fashion Necklines Based on Deep Learning
摘要
Abstract
Accurate detection of fashion design attributes is essential for trend analyses and recommendation systems.Among these attributes,the neckline style plays a key role in shaping garment aesthetics.However,the presence of complex backgrounds and varied body postures in real-world fashion images presents challenges for reliable neckline detection.To address this problem,this research builds a comprehensive fashion neckline database from online shop images and proposes an efficient fashion neckline detection model based on the YOLOv8 architecture(FN-YOLO).First,the proposed model incorporates a BiFormer attention mechanism into the backbone,enhancing its feature extraction capability.Second,a lightweight multi-level asymmetry detector head(LADH)is designed to replace the original head,effectively reducing the computational complexity and accelerating the detection speed.Last,the original loss function is replaced with Wise-IoU,which improves the localization accuracy of the detection box.The experimental results demonstrate that FN-YOLO achieves a mean average precision(mAP)of 81.7％,showing an absolute improvement of 3.9％over the original YOLOv8 model,and a detection speed of 215.6 frame/s,confirming its suitability for real-time applications in fashion neckline detection.关键词
时尚领型检测/深度学习/检测与分类/实时性/YOLOv8Key words
fashion neckline detection/deep learning/detection and classification/real time/YOLOv8分类
轻工纺织引用本文复制引用
陈彩霞,姜琳歆..基于深度学习的时尚领型实时检测方法[J].东华大学学报（英文版）,2025,42(3):301-314,14.基金项目
Fundamental Research Funds for the Central Universities,China(Nos.2232020G-08 and 2232020E-03) （Nos.2232020G-08 and 2232020E-03）
Shanghai University Knowledge Service Platform,China(No.13S107024) （No.13S107024）
