基于改进的YOLOv4输电线路小目标检测

传统的目标检测方法在检测输电线路小目标时,往往存在检测效果不佳,容错率低等问题,针对这种情况,提出一种基于改进的YOLOv4的输电线路小目标检测算法。为了提高输电线路小目标的检测效率,采用一种简化版的YOLOv4算法,减少特征层的使用,从而降低网络计算量。针对输电线路小目标这一特定应用,利用K-means++算法重新进行聚类,得到这一特定场合下的锚点框。实验结果显示,该方法与传统的YOLOv4相比,虽然检测精度有所下降,但是检测速度有大幅度的提高,适用于移动端使用。

Small target detection of transmission line based on improved YOLOv4

Traditional target detection methods often have poor detection effect and low fault tolerance when detecting small target of transmission line.In view of this situation,a small target detection algorithm based on improved YOLOv4 is proposed.In order to improve the detection efficiency of small targets on transmission lines,a simplified version of YOLOv4 algorithm is adopted to reduce the use of feature layer and the amount of network calculation.For the specific application of transmission line small target,K-means++algorithm is used to re cluster to get the anchor box in this specific situation.The experimental results show that,compared with the traditional YOLOv4,the detection accuracy of this method is reduced,but the detection speed is greatly improved,which is suitable for mobile terminal.