基于SEPyconv-DETR的无人机航拍绝缘子检测算法

SEPyconv-DETR-based Drone-assisted Insulator Detection Algorithm

针对绝缘子由于大小不一致、被遮挡等因素导致航拍检测效果不佳的问题,提出了一种基于SEPyconv-DETR的绝缘子检测算法。首先,将Pyconv引入DETR主干网络,有效融合多尺度特征。其次,利用通道注意力SE强化关键特征。然后,将混淆卷积前馈网络(CC-FFN)融合到transformer中,增强局部感知和相邻token之间的联系。最后,综合Alpha-IoU与L1函数构建回归损失,以提高定位精度。实验结果表明,所提算法的检测精度mAP@0.5∶0.95达到了64.1%,优于DETR、YOLOv5等主流算法,并且可有效检测复杂背景下不同尺寸及被遮挡的绝缘子。

In view of the problem of poor detection effect of drone-acquired insulator images due to factors such as inconsistent size and occlusion,an insulator detection algorithm based on SEPyconv-DETR was proposed by the present study.First Pyconv was introduced into the DETR backbone network to effectively fuse multi-scale features.Second the channel attention SE was used to strengthen the key features.Then confusion convolution feedforward network(CC-FFN)was integrated into the transformer to enhance connection between local perception and adjacent tokens.Finally the regression loss is constructed by combining Alpha-IoU and L 1 function to improve the positioning accuracy.The experimental results showed that the proposed algorithm can achieve a detection accuracy of mAP@0.5:0.95 reaching 64.1%and surpassing prevailing algorithms such as DETR and YOLOv5,and can effectively detect insulators with different sizes and occlusions from complex backgrounds.