改进YOLOv5算法的卡通动漫人物脸部检测方法

打开文本图片集

中图分类号：TP391.4 文献标志码：A 文章编号：1007-2683（2025）04-0086-10

A Cartoon and Anime Character Facial Detection Method with Improved YOLOv5 Algorithm

ZHONG Weifeng'， GONG Yue1， LIU Jie 1，2 （1.School of Automation，Harbin University of Science and Technology，Harbin 15Oo80，China; 2.Institute of Automation，Chinese Academy of Sciences，Beijing 1Ooo80，China）

Abstract：Manywell-nowncartoonandanime intellectual properties（IP）havespawnednumerous peripheral productsand culturalifluences，ringingincalulablecommercialvaluetdayIesposetthecurentisseofuauthozeduseofopularae IPsforprofitwithout the creators’consent，weproposeanimprovedcartooncharacter facial detectionmethodbasedon YOLOv5.Firstly，tomettheengineringapplicationrequirementsforeficiencyandlowFLOPs，wereplaceandenhancetheoriginal YOLOv5backbonenetwork，SPDarknet53，withFasteNet，resultiginamoreeficientandlightweightmodel.Secondly，todre theiregularfacialcharacteristicsofcartoonandanimecharacters，weemployanasymmetricweightedbidirectionalfeaturepyramid network（ABiFP）structuretoimprovetheoriginalfeaturefusionmodule，enhancingthenetwork'sabilitytoreceivenon-rectangular receptivefields.Testresultsdemonstratethattheimprovedalgoritmofersincreasdaccuracyreaterodeleficiency，andaigher level of performance.Compared to the previous YOLOv5 model，the improved network achieves an accuracy rate of 90.7% ，a4.6 percentage point increase，and a 55% reduction in network parameters，making it suitable for practical use.

Keywords：cartoonand anime character faces；object detection；model lightweighting；deep learning；YOLOv5 algorithm

0 引言

卡通形象脸部检测作为一种人脸检测技术与动漫卡通产业融合而得到的新兴研究方向，一经提出就受到了广泛的关注与研究。（剩余12587字）