研究面向清代旗鞋装饰纹样的自动化检测,系统评估模型架构、标注策略与优化方法对性能的影响,并探讨任务定义在技术选型中的关键作用。方法上,设计了“宏观整体”与“宏观-分块联合”两种标注方案,分别在不同任务设定下对YOLOv8与YOLOv11系列进行对比评测;同时引入谱空间注意力模块（SSAM）,并结合基于预训练权重的迁移学习开展细粒度类别分析与消融实验。主要结果表明：SSAM在大型模型（YOLOv8x）上带来显著提升,但其收益对基础模型容量高度敏感;最优模型选择依赖于标注策略,在宏观整体任务中,YOLOv8x表现最佳,且YOLOv8系列整体优于同规模的YOLOv11系列;类别层面存在明显差异,其中文字纹样检测相对稳定,而动物纹样识别为主要挑战。研究发现,迁移学习是获取高性能的关键因素,其贡献显著超过单一结构性改进;确立“以任务定义驱动技术选型”的范式,可为文化遗产数字化中的细粒度目标检测提供实证依据与方法参考。
Abstract
This study investigates the automated detection of decorative motifs on Qing Dynasty banner shoes and systematically evaluates the effects of model architectures, annotation strategies, and optimization methods on detection performance, while examining the critical role of task definition in guiding technology selection. Two annotation schemes, namely "macro-level holistic" and "macro-patch joint", were designed, and comparative experiments were conducted on the YOLOv8 and YOLOv11 series under different task settings. In addition, a spectral-spatial attention module (SSAM) was introduced, and transfer learning based on pre-trained weights was employed to perform fine-grained category analysis and ablation experiments. The results show that SSAM yields significant improvements on large-scale models (YOLOv8x), although its effectiveness is highly sensitive to the capacity of the base model. The optimal model selection depends on the annotation strategy. Under the macro-level holistic task setting, YOLOv8x performs the best, and the YOLOv8 series generally outperforms the YOLOv11 series of comparable scale. At the category level, notable performance differences are observed, with textual motifs being detected relatively reliably, while animal motifs remain the primary challenge. Furthermore, the study finds that transfer learning is a key factor in achieving high performance, with its contribution significantly exceeding that of individual structural modifications. These findings establish a paradigm of "task-definition-driven technology selection", providing empirical evidences and methodological references for fine-grained object detection in cultural heritage digitization.
关键词
清代旗鞋 /
标注策略 /
YOLOv8 /
YOLOv11 /
谱空间注意力模块（SSAM）
Key words
Qing Dynasty banner shoes /
annotation strategy /
YOLOv8 /
YOLOv11 /
spectral-spatial attention module (SSAM)
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] 定宜庄. 清末民初的“满洲”“旗族”和“满族”[J]. 清华大学学报(哲学社会科学版),2016,31(2):91-98.
[2] 卢彦梅,陈明东,史宇亮,等. 基于改进YOLOv8n的黄瓜叶片病害识别[J]. 中国农机化学报,2026,47(2):94-100.
[3] 郝敏,汤宇翔,孙建英,等. 基于改进YOLOv8的马铃薯种薯芽眼检测[J]. 中国农机化学报,2026,47(2):63-70.
[4] 邵自强,魏利胜,武涛. 改进YOLOv11n的高效交通实例分割算法[J]. 电子测量与仪器学报,2026,40(2):95-106.
[5] 杨超凡,张仕林,戴飞,等. 基于改进YOLOv11n的玉米根茬检测算法[J]. 农业工程学,2026,42(1):171-180.
[6] 满懿. 中国少数民族服饰文化与传统技艺-满族[M]. 北京:中国纺织出版社,2022:229.
[7] TANG L D,YUN L J,CHEN Z Q,et al.HRYNet:A Highly Robust YOLO Network for Complex Road Traffic Object Detection[J]. Sensors,2024,24(2):642.
[8] 谢明宏. 基于改进YOLOv8的无人机视角下小目标检测模型[J]. 信息技术与信息化,2025(3):7-11.
[9] ALI TUTUNCUOGLU R,GUNEY S.Development of "Target Detection and Identification System" for Combat Military Vehicles via Artificial Intelligence and Decision Support Matrix[J]. International Journal for Multidisciplinary Research,2024,6(6):34237.
[10] AFIFAH V,ERNIWATI S.YOLOv8 for Object Detection:A Comprehensive Review of Advances,Techniques,and Applications[J].:International Journal of Advanced Computing and Informatics,2025,2(1):53-61.
[11] SAPKOTA R,FLORES-CALERO M,QURESHI R,et al.YOLO Advances to Its Genesis:A Decadal and Comprehensive Review of the You Only Look Once (YOLO) Series[J]. Artificial Intelligence Review,2025,58(9):274.
[12] LI Q,XU B Y,ZHU C H,et al.Improved YOLOv8-Based Marine Life Detection Algorithm[J]. Journal of Electronic Research and Application,2025,9(1):7-13.
[13] LU X J.Research on Multi-Feature Fusion of Improved Concat Module Based on Yolov11 in Wheat Spike Recognition[J]. International Journal of Cognitive Informatics and Natural Intelligence,2025,19(1):1-18.
[14] ZHOU K Q,JIANG S H.Forest Fire Detection Algorithm Based on Improved YOLOv11n[J]. Sensors,2025,25(10):2989.
[15] FU Q,LUO Y,WU A.MAS-YOLOv11:An Improved Underwater Object Detection Algorithm Based on YOLOv11[J]. Sensors,2025,25(11):3433.
[16] 侯旺旺. YOLOv11技术在学生注意力识别中的应用[J]. 信息记录材料,2025,26(5):121-123.
[17] TIAN Z,YANG F,YANG L,et al.An Optimized YOLOv11 Framework for the Efficient Multi-Category Defect Detection of Concrete Surface[J]. Sensors,2025,25(5):1291.
[18] SETYAWAN WIJAYA B,MUNIR R,UTAMA N P.TerraSegNet:Bilateral Axial Attention Network for Remote Sensing Image Segmentation in Diverse Environmental Monitoring Applications[J].IEEE Access,2025,13:208868-208901.
基金
2024年度国家哲学社会科学（艺术学）青年项目（24CG215）; 2019年度国家哲学社会科学（艺术学）重大项目（19ZD23）
PDF(10117 KB)
渝公网安备50010702505643 渝ICP备15012534号-8
