基于EfficientDet的输电线路破损防振锤检测

EfficientDet-based detection of damaged vibration damper in transmission line

随着人工智能尤其是深度学习技术的快速发展,无人机巡检与图像识别技术在电网输电线路破损防振锤检测中发挥着重要作用。文中构建了一种基于改进EfficientDet深度神经网络模型的破损防振锤检测模型。采用先目标检测后分类判别的方法实现输电线路中的破损防振锤识别,基于目标各要素之间的相互关系判别技术,优化了背景干扰所产生的误识别问题。使用细节特征提取来判断拍摄倾角并去除倾斜角度过大的目标。实验结果表明,文中所提改进EfficientDet目标检测模型的mAP为51.16%,准确率与召回率分别为93.3%、91.8%,均优于其他目标检测模型。同时,破损防振锤的分类准确率与召回率分别达到85.4%、81.7%,由此验证了所提方法的准确性与实用性。

With the rapid development of artificial intelligence,especially deep learning technology,UAV inspection and image recognition technology play an increasingly important role in the detection of transmission line damaged damper.In this paper,a two-step detection model of damaged damper based on improved EfficientDet depth neural network model is constructed.The method of target detection before classification is used to identify the damaged damper in the transmission line.Based on the relationship between the target elements,the problem of false identification caused by background interference is optimized.The method of using detail feature extraction to determine the shooting inclination angle is used to remove the target with too large inclination angle.The experimental results show that the mAP of the improved EfficientDet target detection model is 51.16%,the accuracy and recall are 93.3% and 91.8%respectively,which is better than other target detection models.At the same time,the classification accuracy and recall rate of damaged damper are 85.4% and 81.7% respectively,which verify the accuracy and practicability of the proposed method.