Table 6 Quantitative comparison of performance with 8 currently advanced methods based on the dataset.
Algorithm | Backbone | AP (0.5)% | ||||
|---|---|---|---|---|---|---|
Damage | Flashover | Shed missing | Insulator | mAP (0.5)% | ||
RetinaNet (2017) | R-50-FPN | 76.3 | 77.5 | 84.3 | 93.6 | 82.93 |
DETR (2020) | R-50 | 77.4 | 75.9 | 88.2 | 94.5 | 84.00 |
Mask R-CNN-Swin | SwinTransformer | 78.3 | 75.3 | 89.6 | 91.3 | 83.63 |
Sparse R-CNN (2021) | R-50-FPN | 58.2 | 70.9 | 91.3 | 93.8 | 78.55 |
YOLOX (2021) | YOLOX-s | 68.3 | 68.3 | 90.2 | 93.2 | 80.00 |
RTM-Det (2022) | RTMDet-s | 74.2 | 74.6 | 89.6 | 94.6 | 83.25 |
YOLOv7 (2023) | CSPDarknet-53 | 69.9 | 78.5 | 90.8 | 91.3 | 82.63 |
YOLOv7 + MCI-GLA (2024) | CSPDarknet-53 | 75.6 | 76.3 | 93.6 | 95.4 | 85.23 |
Ours | CSPDarknet-53 | 79.7 | 80.5 | 94.9 | 96.1 | 89.90 |
