改进型YOLO-V4模型的电力杆塔状态评估探索

Exploration of power tower condition assessment based on improved YOLO-V4 model

针对电网线路存在倒杆、断杆现象以及现有技术巡检方式落后的问题,提出一种新型的电力杆塔状态评估方法,构建出能够实现电力杆塔位置定位的评估方法,并引入电力杆塔检测的YOLO-V4模型深度学习算法,该算法模型包括53个卷积层,具有大量的3*3、1*1的卷积核,该算法还具有Darknet-53特征提取网络、多尺度融合特征网等,通过评价函数对所应用的YOLO-V4目标检测网络的损失进行检测。试验表明,YOLO-V4模型深度学习算法引入GIoU指标后,相比普通状况平均精度(AP)从97.12%提高到98.94%,准确率(Precision)从94.5%提高至95.6%,召回率(Recall)从97.5%提高至99.2%。

In order to solve the problems of inverted poles and broken poles in power grid lines and backward inspection methods in the prior art,a new method for evaluating the status of power poles and towers is proposed,and an evaluation method that could realize the location of power poles and towers is constructed.YOLO-V4 model deep learning algorithm which includes 53 convolutional layers with a large number of 3*3,1*1 convolution kernels in it is introduced,and the algorithm also has Darknet-53 feature extraction network,multi-scale fusion feature network,etc.The loss of YOLO-V4 target detection network is detected by the evaluation function.Experiments show that the YOLO-V4 model deep learning algorithm in this study introduces the GIoU index,and its(AP)increases from 97.12%to 98.94%,accuracy(Precision)increases from 94.5%to 95.6%,recall rate(Recall)increases from 97.5%to 99.2%compared with the ordinary situation average accuracy.