An Image-Based Deep Learning Approach with Improved DETR for Power Line Insulator Defect Detection

Figure 11

The average precision curves of IoU50, 75, and 50 : 95. (a) is the results of data set “CPLID,” a synthetic composite insulator data set, and (b) is from data set “B,” which is a data set containing real-world glass insulator images collected by us. As we can see from the curves, the AP75 and AP50:95 of real-world data dropped a lot compared to the synthetic, and the AP50 curve converges to a competitive level.

(a)

(b)