基于上下文增强与多目标语义感知的电力缺陷检测算法

胡欣,阎希玥,常娅姝,程鸿亮,肖剑,罗诗伟,马亮

Context-enhanced multi-target semantic perception algorithm for power defect detection

Xin HU,Xiyue YAN,Yashu CHANG,Hongliang CHENG,Jian XIAO,Shiwei LUO,Liang MA

表 8 核心模块与其他主流方法对比

Tab.8 Comparison of core modules and mainstream methods

模型 P R mAP50 mAP50-95 先进骨干 网络 StartNet 0.682 0.548 0.587 0.438 Efficient Formerv2 0.638 0.528 0.558 0.412 HGnetV2 0.696 0.569 0.607 0.449 Ghost HGNetV2 0.676 0.565 0.600 0.446 PKINet 0.632 0.526 0.558 0.414 c2f-PKI 0.687 0.541 0.591 0.448 LDDM-ResNet 0.701 0.578 0.616 0.466 先进注意力 机制 Efficient Additive 0.689 0.549 0.605 0.457 Cascaded Group 0.679 0.564 0.615 0.468 HilO 0.689 0.549 0.601 0.460 LPE 0.690 0.558 0.606 0.466 DA-Encoder 0.687 0.575 0.616 0.474 先进下采样 模块 CG-Down 0.665 0.592 0.613 0.470 HSFPN 0.676 0.576 0.608 0.459 HSPAN 0.659 0.538 0.576 0.438 Waveletpool 0.584 0.463 0.485 0.364 slimneck 0.686 0.568 0.601 0.451 LGCA 0.702 0.575 0.628 0.482