融合共现推理的Faster R-CNN输电线路金具检测

为促进目标检测模型与电力领域业务知识有机融合，缓解金具样本间长尾分布问题，有效提升输电线路金具的自动化检测效果，提出了融合共现推理的Faster R-CNN(faster region-based convolutional neural network)输电线路金具检测模型。首先，深入研究输电线路金具结构化组合规则，通过数据驱动的方式以条件概率对金具目标间的共现连接关系进行有效表达；然后，结合图学习方法，利用学习并映射的共现概率关联作为共现图邻接矩阵，使用基线模型(Faster R-CNN)提取的特征向量作为图推理输入特征，以自学习的变换矩阵作为共现图传播权重，完成图信息传播并实现有效的共现推理模型嵌入。实验证明，融合共现推理模块的Faster R-CNN模型较原始模型提升了6.56%的准确率，对于长尾分布样本中数量较少的金具性能提升尤其显著。

Integrating co-occurrence reasoning for Faster R-CNN transmission line fitting detection

To promote the organic integration of object detection and business knowledge in the electric power field, alleviate the problem of long-tailed distribution among fitting samples, and effectively improve the automatic detection effect of transmission line fittings, we propose a faster region-based convolutional neural network (Faster R-CNN) transmission line fitting detection model based on integrating co-occurrence reasoning. First, the structural combination rule of transmission line fittings is extensively investigated, and the co-occurrence connection relationship between objects is effectively expressed using conditional probability in a data-driven manner. Then, in combination with the graph learning method, the co-occurrence probability association is learned and mapped as the adjacency matrix of the co-occurrence graph, the feature vector extracted from the baseline model (Faster R-CNN) is used as the graph inference input feature, and the self-learning transformation matrix is used as the propagation weight of the co-occurrence graph to complete graph information propagation and realize effective co-occurrence inference model embedding. The experimental results show that the Faster R-CNN integrating co-occurrence reasoning module outperforms the original model by 6.56%, which is particularly significant for performance improvement in terms of transmission line fitting detection with a reduced long-tailed distribution among fitting samples.