Dual-Feature Fusion Recognition of Wear in Transmission-Line Connecting Fittings Based on the Faster-RCNN Algorithm
摘要
为应对复杂野外环境下输电线路连接金具磨损识别中的图像质量退化及特征提取能力不足的挑战,提出基于 Faster-RCNN 算法的输电线路连接金具磨损双特征融合识别方法.采集输电线路连接金具图像,通过构建大气散射模型提升图像画质,提取图像的梯度方向直方图(HOG)特征(反映边缘与形状信息)和局部二值模式(LBP)特征(反映局部纹理信息),采用交叉注意力机制强化双特征间的信息交互,实现深层特征与表层纹理特征的有效融合,将融合特征输入 Faster-RCNN 模型,通过区域建议网络生成候选区域,经感兴趣区域(RoI)池化层和全连接层完成磨损分类与定位.实验证明该方法在 10 个样本测试中的识别结果与实际磨损情况完全一致,磨损概率值(95.21%、87.68%、65.72%)与严重、中度、轻度磨损等级匹配合理,一致性指数稳定在 0.9 以上,有效建立了输电线路连接金具高精度磨损识别体系,为输电线路运维检测提供可靠技术支撑.
To address image quality degradation and insufficient feature extraction capability in the recognition of wear in transmission line con-necting fittings under complex field conditions,a dual feature fusion recognition method based on the Faster Region-based Convolutional Neural Network(Faster-RCNN)was proposed.Images of transmission line connecting fittings were collected,and an atmospheric scattering model was constructed to improve image quality.Histogram of oriented gradients(HOG)features,representing edge and shape information,and local bi-nary pattern(LBP)features,representing local texture information,were extracted.A cross attention mechanism was used to enhance informa-tion exchange between the two feature types and effectively fuse deep features with surface texture features.The fused features were input into the Faster-RCNN model.Candidate regions were generated by a region proposal network(RPN).Wear was then classified and localized using a region of interest(RoI)pooling layer and fully connected layers.Experiments on 10 samples showed that the recognition results were fully consistent with the actual wear conditions.Wear probabilities of 95.21%,87.68%,and 65.72%corresponded appropriately to severe,moder-ate,and mild wear,respectively,and the consistency index remained above 0.9.The method therefore effectively established a high accuracy wear recognition system for transmission line connecting fittings.It could provide reliable technical support for transmission line operation and maintenance inspection.