结合近景摄影测量与CNA-YOLOv8的降落伞线迹缺陷检测
DOI:
CSTR:
作者:
作者单位:

1.南京信息工程大学遥感与测绘工程学院南京210044;2.南京信息工程大学自动化学院南京210044; 3.北京空间机电研究所北京100094

作者简介:

通讯作者:

中图分类号:

TP391.4;TN911.73

基金项目:

航天进入减速与着陆技术实验室开放基金(EDL19092303)项目资助


Combining photogrammetry with CNA-YOLOv8 for parachute line defect detection
Author:
Affiliation:

1.School of Remote Sensing and Geomatics Engineering, Nanjing University of Information Science and Technology, Nanjing 210044, China;2.School of Automation, Nanjing University of Information Science and Technology, Nanjing 210044, China;3.Beijing Institute of Space Mechanics & Electricity, Beijing 100094, China

Fund Project:

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 资源附件
  • |
  • 文章评论
    摘要:

    降落伞线迹缺陷检测是确保降落伞安全性和可靠性的关键任务。针对降落伞线迹样本中缺陷的定位难度大、类别尺度差异显著及样本数量有限且分布失衡的问题,提出了一种改良YOLOv8算法,命名为CNA-YOLOv8(coordinate attention & non-local pyramid adaptive threshold focal YOLOv8)。首先,该算法在YOLOv8基线模型的主干部分添加坐标注意力机制(coordinate attention, CA),通过将位置信息融入通道注意力,增强模型对线迹结构与局部特征的感知能力。其次,为进一步提升多尺度缺陷的检测性能,设计了非局部金字塔检测头(non-local pyramid detect, NLPD),使单层特征图具备自适应调整感受野的能力,从而有效捕捉不同尺度下的缺陷目标。最后,引入自适应阈值焦点损失函数(adaptive threshold focal loss, ATFL),动态调整困难样本的损失权重,以缓解类别不平衡对模型训练的影响。此外,为解决数据摄取过程中线迹缺陷样本失真的问题,结合近景摄影测量原理,采用ArUcoTag标定板对摄取图像进行正视校正,消除透视畸变,制作出一套供深度学习模型检测的降落伞线迹缺陷数据集。在该数据集上,将CNA-YOLOv8与多种主流检测模型进行对比实验,实验结果表明,CNA-YOLOv8模型在平均精度均值(mAP)mAP@0.5上达到了83.324%,显著优于其他对比模型,凸显出其在降落伞线迹缺陷检测任务中的优越性。同时,该模型相较原始YOLOv8的参数量仅增加1.59%,在检测性能与模型复杂度之间取得了良好平衡,验证了该方法的有效性与实用性。

    Abstract:

    Parachute line defect detection is a critical task for ensuring the safety and reliability of parachutes. Aiming at the challenges in parachute line samples, including the difficulty in defect localization, significant differences in category scales, limited sample quantity, and imbalanced sample distribution, this study proposes an improved YOLOv8 algorithm, named CNA-YOLOv8 (coordinate attention & non-local pyramid adaptive threshold focal YOLOv8). First, the coordinate attention (CA) mechanism is incorporated into the backbone of the baseline YOLOv8 model. By embedding positional information into channel attention, it enhances the model’s perception of stitch structures and local features. Second, to further improve the detection performance of multi-scale defects, a non-local pyramid detect (NLPD) head is designed, enabling the single-layer feature map to adaptively adjust the receptive field and thus effectively capture defect targets at different scales. Finally, an adaptive threshold focal loss (ATFL) is introduced to dynamically adjust the loss weights of hard samples, mitigating the impact of class imbalance on model training. In addition, to address the distortion of stitch defect samples during data acquisition, this study combines the principle of close-range photogrammetry and uses ArUcoTag calibration boards to perform orthographic correction on captured images, eliminating perspective distortion and constructing a dedicated parachute stitch defect dataset for deep learning model detection. Comparative experiments between CNA-YOLOv8 and several mainstream detection models are conducted on this dataset. Experimental results show that the proposed CNA-YOLOv8 achieves a mean average precision (mAP) mAP@0.5 of 83.324%, significantly outperforming other comparative models, highlighting its superiority in parachute stitch defect detection. Meanwhile, the number of parameters of this model only increases by 1.59% compared with the original YOLOv8, achieving a favorable balance between detection performance and model complexity, which verifies the effectiveness and practicality of the proposed method.

    参考文献
    相似文献
    引证文献
引用本文

冯欢,严清赟,陈禹汗,张菀,刘靖雷,杨霞.结合近景摄影测量与CNA-YOLOv8的降落伞线迹缺陷检测[J].电子测量与仪器学报,2026,40(9):163-177

复制
分享
相关视频

文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:
  • 最后修改日期:
  • 录用日期:
  • 在线发布日期: 2026-11-13
  • 出版日期:
文章二维码
×
《电子测量与仪器学报》
关于防范虚假编辑部邮件的郑重公告