基于深度学习的输电线路销钉缺陷检测

Transmission Line Pin Defect Detection Based on Deep Learning

销钉是输电线路中的重要器件,起到固定螺母的作用,一旦出现故障,可能造成大面积停电等隐患。由于销钉缺陷目标较小,对于目标检测算法来说是一种很大的挑战。针对特征金字塔算法做出了改进,提出PinNet。首先,特征提取层换为残差网络的最新变体SCNet,提取更具辨别力的特征。其次,在FPN的基础上,设计了PinFPN,进一步增强了底层的语义信息和位置信息,提高检测小目标的能力。最后,针对数据集不足的问题,采用多种方式的数据增强,扩充训练样本,提高模型的鲁棒性。在测试集上,相比于原算法,该模型在销钉缺失目标上的AP值提高了4.2%,与其他主流的算法相比,也具有很大的优势。为了进一步证明算法的有效性,在其他输电线路小目标缺陷数据集上测试,同样可以取得不错的效果。

Pins,the important components in the transmission lines,functions to fix the nuts.Once a pin defect occurs,a power failure in a large area may be caused.The tiny size of a pin makes the pin defect detection a challenging task.Therefore,we improve the Feature Pyramid Network and propose a better network structure specifying for pin defect detection--PinNet.First,the feature extraction layer is replaced by SCNet,the latest variant of the residual network,to extract more discriminative features.Second,on the basis of FPN,PinFPN is designed to further enhance the underlying semantic information and location information,and also to improve the ability to detect small objects.Finally,to solve the problem of lack of dataset,various methods of data augmentation are adopted to expand the training samples and improve the robustness of the model.On the test set,compared with the original algorithm,the AP value of our model on the pin missing target is improved by 4.2%,which also has a great advantage compared with other mainstream algorithms.In order to further prove the effectiveness of the algorithm,the model has been tested on the small target defect data sets of other transmission lines,which also achieves good results.