基于注意力特征融合的漏磁缺陷识别方法

Magnetic flux leakage defect recognition method based on attention feature fusion

针对漏磁缺陷识别率低、检测速度慢等问题,提出了一种基于注意力特征融合的漏磁缺陷识别方法。所提算法以CenterNet为基础进行修改,主干网络选取了一种轻量级网络PP-LCNet,相较于现在流行的主干特征提取网络既保证了低计算量又保证了高精度。采用注意力网络CBAM主动学习低层特征中的重要信息并与高层特征进行融合,使模型同时获得低层细粒度信息与高层语义信息,进而提升小缺陷识别的准确率。结果表明,当IOU大于0.5时,所提算法的准确率为94.3%,推理时间为9.6 ms。

Aiming at the problem of low recognition rate and slow detection speed of magnetic flux leakage(MFL)defects,a MFL defect recognition method based on attention feature fusion was proposed.The algorithm was modified on the basis of CenterNet.A lightweight network PP-LCNet was selected as the backbone network,which could simultaneously guarantee low computation and high accuracy,compared with the popular backbone feature extraction network.At the same time,the attention network CBAM was used to positively learn the important information about the low-level features and integrate it with the high-level features.The model could obtain both low-level fine-grained information and high-level semantic information to improve the accuracy of small defect recognition.The experimental results show that the accuracy of as-proposed algorithm is 94.3% and the inference time is 9.6 ms,respectively,when the IOU is greater than 0.5.