王满利，杨爽，张长森，等.基于改进YOLOv8n的立井刚性罐道接头错位检测算法［J］.河南理工大学学报（自然科学版），doi:10.16186/j.cnki.1673-9787.2023090041

WANG M L, YANG S, ZHANG C S，et al. An improved YOLOv8n based dislocation detection algorithm for shaft rigid tank passage joint［J］.Journal of Henan Polytechnic University（Natural Science），doi:10.16186/j.cnki.1673-9787. 2023090041

基于改进YOLOv8n的立井刚性罐道接头错位检测算法(网络首发)

王满利¹，杨爽¹，张长森¹，张善文²

（1.河南理工大学 物理与电子信息学院，河南 焦作 454000；2. 西京学院 电子信息学院，陕西 西安 710000）

摘要: [目的] 立井刚性罐道作为提升系统的导向装置，对于提升系统的安全稳定运行至关重要。然而在使用过程中，罐道很容易发生横向移位和变形，一旦发生，就会引起提升系统的强烈振动，对其安全运行造成严重影响。为及时发现立井刚性罐道接头故障，消除提升系统运行隐患，提出一种基于改进YOLOv8n的立井刚性罐道接头错位检测算法YOLOv8n-CSS。[方法] 首先，为了避免井筒中采集的图像特征细节弱与不均衡，将CA注意力机制模块融入YOLOv8n主干网络，融合通道和空间2个维度关系位置信息，有助于模型更加集中地关注输入图像的重要部分，提高模型的识别准确性和泛化能力，提高YOLOv8n主干网络在模糊、黑暗、强光环境下对局部重要信息特征提取能力；其次，为适应不同尺度的输入图像，使网络获取更全面的细节特征信息，改进SPPF模块提高网络的性能；再次，为了降低回归自由度，并进一步提高回归精度，使用SIOU损失函数，提高锚框预测精度；最后，采用非定焦测距法计算罐道接头偏移尺寸。[结果] 结果表明，和YOLOv8n网络相比，YOLOv8n-CSS网络在立井刚性罐道接头数据集上，精度P提升了2.4%，召回率R提升了15.7%，平均精度由YOLOv8n的83.8%提升到94.6%，提高了10.8‬%。[结论] 相比其他YOLO算法，在立井刚性罐道接头错位检测中本文改进算法具有显著的优势。

关键词: 刚性罐道；接头错位；YOLOv8n；CA注意力机制；改进SPPF模块

中图分类号：TP391.4

doi: 10.16186/j.cnki.1673-9787.2023090041

基金项目: 国家自然科学基金资助项目(62172338)； 河南省科技攻关项目（242102221006）

收稿日期：2023-09-18

修回日期：2024-03-08

网络首发日期：2025-04-27

An improved YOLOv8n based dislocation detection algorithm for shaft rigid tank passage joint

WANG Manli¹, YANG Shuang¹, ZHANG Changsen¹, ZHANG Shanwen²

(1.School of Physics ＆ Electronic Information Engineering，Henan Polytechnic University，Jiaozuo 454000，Henan, China；2. School of Electronic Information, Xijing University, Xi’an 710000, Shaanxi, China)

Abstract: [Objective] As the guiding device of the lifting system, the rigid shaft channel is very important for the safe and stable operation of the lifting system. However, in the process of use, the tank passage is prone to lateral displacement and deformation; once the passage is uneven, it will cause strong vibration of the lifting system, which has a serious impact on the operation and safety of the lifting system. In order to detect the faults of the rigid tank channel joints of vertical wells in time and eliminate the hidden dangers of the operation of the hoisting system, a misalignment detection algorithm of the rigid tank channel joints of vertical wells based on the improved YOLOv8n, YOLOv8n-CSS, is proposed. [Methods] First of all, in order to suppress the interference of weak and unbalanced image feature details collected in the wellbore, The CA attention mechanism module was integrated into the YOLOv8n backbone network to integrate the location information of the two dimensions of channel and space, which helped the model pay more attention to the important parts of the input image, improve the recognition accuracy and generalization ability of the model, and improve the feature extraction ability of local important information of the YOLOv8n backbone network under fuzzy, dark, and strong light environments. Then, in order to adapt to the input images of different scales, the network can obtain more comprehensive, detailed feature information, and the SPPF module is improved to improve the performance of the network. Then, in order to reduce the degree of freedom of regression and further improve the accuracy of regression, the SIOU loss function is used to improve the prediction accuracy of the anchor frame. Finally, the offset size of the channel joint is calculated by using the non-fixed focus ranging method. [Results] The results show that compared with the YOLOv8n network, the accuracy P and recall rate R of the YOLOv8N-CSS network on the data set of shaft rigid channel joints increased by 2.4%, an increase of 15.7%, and an average accuracy increase of 10.8% from 83.8% of the YOLOv8n network to 94.6%. [Conclusion] Compared with other YOLO algorithms, the improved algorithm in this paper has significant advantages in the dislocation detection of rigid channel joints in shafts.

Key words: rigid guide; joint dislocation; YOLOv8n; CA attention mechanism; improved SPPF module

CLC: