融合迁移学习与光照自适应的轻量化YOLOv11竹节检测优化模型研究

打开文本图片集

中图分类号：TB9：G232 文献标志码：A 文章编号：1674-5124（2025）09-0176-07

Abstract： Accurate and robust detection of bamboo joints is essential for automated bamboo processing， yet current methods face challnges such as low detection accuracy and unstable coordinate localization under varying lighting conditions. Thisstudy proposes an optimized lightweight YOLOvl1-based detection framework that integrates transfer learning and an illumination-adaptive mechanism. The approach leverages a pre-trained model on a wood defect dataset to transfer domain knowledge to bamboo joint detection， addressing the limitations of limited annotated bamboo data.Furthermore，an illumination-adaptive detection threshold is introduced to enhance robustness under diverse lighting environments encountered in industrial setings.Experimental evaluation on the publicly available Roboflow bamboo joint dataset demonstrates that the proposed model achieves an mAP@0.5 of 99.3% ，representinga O.3 percentage point improvement over the baseline YOLOvlln model. The model also exhibits enhanced stability across varying lighting conditions. The results indicate that the proposed method provides a practical solution for improving the accuracy and reliability of bamboo joint detection in automated processing systems without increasing training costs.

Keywords： transfer learning; YOLOv11; bamboo joint detection; ilumination-adaptive threshold; lightweight detection

0 引言

在竹材产业中，竹材初加工工序主要包括锯竹、分选、剖竹、开片、去青去黄去节、初剖及精剖等环节。（剩余9774字）