基于改进YOLOX的变电站设备缺陷检测方法

Defect Detection for Substation Based on Improved YOLOX

为减轻电力工作人员的巡检负担,实现变电站智能巡检,对变电站设备缺陷检测算法进行了研究。首先,利用数据增强方法对有限的初始数据集进行扩充,利用多种图像处理方法增加数据集的复杂度,生成考虑复杂光照环境的数据集;然后,采用自适应空间特征融合(ASFF:Adaptively Spatial Feature Fusion)的方法缓解特征金字塔中不同尺度特征的不一致性问题,并引入Focal损失函数作为置信度损失函数以缓解正负样本不平衡的问题,利用改进的YOLOX-s(You Only Look Once X-s)网络模型设计了变电站缺陷检测算法;最后,将改进的YOLOX-s网络模型与其他深度学习算法的检测效果进行对比,实验结果表明,改进的YOLOX-s网络模型的综合检测效果较好,准确性和实时性均可以满足变电站设备缺陷检测任务。

In order to reduce the inspection burden of electric power workers and realize intelligent inspection in substation,the algorithm of substation equipment defect detection is studied.Firstly,the data augmentation method is used to expand the initial dataset and various image processing method is used to generate the dataset with complex illumination environment.Then,the adaptive spatial feature fusion method is used to mitigate the inconsistency of different scale features in the feature pyramid,and the loss function of confidence is changed to Focal loss function to mitigate the imbalance between positive and negative samples.Based on the improved YOLOX-s(You Only Look Once X-s) network model,the algorithm of substation defect detection is designed.Finally,the detection effect of the improved YOLOX-s model is compared with that of other deep learning algorithms.Under the designed data set,the experiment shows that the comprehensive detection effect of the improved YOLOX-s network model is good,and the accuracy and real-time performance is satisfied.