目的 螺栓是输电线路中数量最多的紧固件,一旦出现缺陷就会影响电力系统的稳定运行。针对螺栓缺陷自动检测中存在的类内多样性和类间相似性挑战,提出了一种融合先验信息和特征约束的Faster R-CNN模型训练方法。方法 在航拍巡检图像预处理阶段,本文设计了基于先验信息的感兴趣区域提取算法,能够提取被识别目标的上下文区域,从而减少模型训练阶段的数据量,帮助模型在训练阶段关注重点区域,提高其特征提取能力。在模型训练阶段,首先通过费舍尔损失约束Faster R-CNN模型的输出特征生成,使样本特征具有较小的类内距离和较大的类间间隔；然后采用K近邻算法处理样本特征得到K近邻概率,将其作为难易样本的指示以引导模型后续更加关注难样本。结果 本文在真实航拍巡检图像构建的螺栓数据集上进行测试,与基线模型相比,本文模型使螺栓识别的平均精度均值(Mean Average Precision, mAP)提高了6.4%,其中正常螺栓识别的平均精度(Average Precision, AP)提高了0.9%,缺陷螺栓识别的平均精度提高了12%。结论 本文提出的融合先验信息和特征约束的输电杆塔螺栓缺陷检测方法在缺陷螺栓识别上获得了良好的效果,为实现输电线路螺栓缺陷的自动检测奠定了良好的基础。

Objective Bolts, as fasteners, play a role in connecting and fixing key components in power tower. Adverse environment, mechanical vibration and material aging can cause bolt cotter pins to fall off or come out, affecting the normal operation of other parts on the transmission line. In order to ensure the uninterrupted operation of power system, the monitoring, testing and maintenance of bolt condition play a crucial role. It is a difficult problem in power system how to realize automatic detection of transmission line bolt defects and timely and efficiently troubleshoot faults. Aiming at the challenges of intra-class diversity and inter-class similarity in automatic bolt defect detection, a Faster R-CNN model training method based on prior information and feature constraints was proposed. Method In the pre-processing stage of aerial inspection image, this paper designs a region of interest extraction algorithm based on prior information, which can extract the context region of the identified object, so as to reduce the amount of data in the training stage, help the model to focus on the key areas in the training stage, and improve its feature extraction ability. In the model training stage, firstly, the output features of Faster R-CNN model are constrained by Fisher constraint to make the sample features have smaller intra-class distances and larger inter-class intervals. Then the k-nearest neighbor algorithm is used to process the sample features to get the k-nearest neighbor probability, which is used as an indication of difficult samples to guide the model to pay more attention to difficult samples in the future. Result All the data used in this paper are real aerial inspection images from a power company in Central China. Bolts on the power tower can be divided into bolts with and without pins. Bolt with pin plays the role of connecting and fixing key components, which needs to bear a large force. Compared with bolt without pin, bolt with pin is more prone to defects. Therefore, the object of this paper is bolt with pin. According to the state of the cotter pin on the bolt, the bolt with pin is divided into normal bolt and defective bolt, wherein the defective bolt refers to the bolt whose cotter pin falls off or loosens. There are 28887 images in the data set, and the resolution of each image is about 5000×3000 pixels. The training set, validation set and test set are divided according to the ratio of 8:1:1. The proposed model is tested on this dataset. Compared with the baseline model, the Mean Average Precision (mAP) of bolt recognition is improved by 6.4%. The Average Precision (AP) of normal bolt identification is increased by 0.9%, and the average accuracy of defective bolt identification is increased by 12%. Conclusion The proposed method based on prior knowledge and feature constraints achieves good results in the identification of defective bolts, which lays a good foundation for the automatic detection of transmission line bolt defects.