结合KL散度和形状约束的Faster R-CNN典型金具检测方法

Typical Fittings Detection Method with Faster R-CNN Combining KL Divergence and Shape Constraints

输电线路航拍巡检图像中典型金具的高准确率检测是其状态检测和故障诊断的基础。针对金具与金具之间、金具与背景之间相互干扰导致模型出现的目标检测得分很高但检测框位置存在极大误差的问题,提出了一种结合KL(Kullback-Leibler)散度和形状约束的Faster R-CNN(faster region-based convolutional neural netwoks)典型金具检测方法,在Faster R-CNN检测模型框架的基础上,增加对目标边界框分布的预测,同时使用KL散度度量金具坐标预测分布与真值分布之间的距离,并将其作为损失函数优化模型参数,进一步将数据集中不同金具类别目标的形状特征作为约束加入损失函数中,以提高模型的泛化性能和边界框回归精度。实验结果证明:提出的方法在一定程度上解决了检测模型目标边界框回归不准确的问题,其中,各类别平均准确率的均值(mean average precision,mAP)由79.76%提高到了83.68%。研究可为进一步对典型金具进行状态检测和故障诊断奠定基础。

High-accuracy detection of typical fittings in aerial inspection images of power transmission line is the basis for their status detection and fault diagnosing.Aiming at the phenomenon that the model has a high object detection score but a large error in the detection position due to the complicated positional relationships between the fittings and the fittings,and between the fittings and the background,we put forward a typical fittings detection method based on faster region-based convolutional neural networks(Faster R-CNN)that combines Kullback-Leibler(KL)divergence and shape constraints.Based on the Faster R-CNN detection model,a prediction branch for the objects bounding box distribution is added.And then,KL divergence is used to measure the distance between the predicted distribution of the fittings coordinates and the ground truth distribution.It is used as a loss function for optimizing model parameters.Furthermore,the shape features of different fitting categories in the data set are used for constraining the loss function,which improves the generalization performance of the model and the accuracy of bounding box regression.The experiments prove that the proposed method can be adopted to solve the problem of inaccurate regression of the bounding box to a certain extent,and the mean average precision(mAP)value is increased from 79.76%to 83.68%.The study has laid foundation for further condition monitoring and fault diagnosis of typical fittings.