Fig. 8 Predicted results of the improved RetinaNet for
Si3N4 ceramic bearing ring surface
defect dataset
Using the improved RetinaNet model for the verification set of defects
on the surface of Si3N4 ceramic
bearings, the prediction results are shown in Figure 8. The red, green,
purple and blue boxes in the figure represent the prediction boxes for
pit, Wear, crack and snowflake, respectively. The prediction results of
some defect images are shown in this figure. The prediction values of
Pit are 0.98, 0.95, 0.99, and 0.98. The prediction values of wear are
1.00, 1.00, 0.99, and 0.99. The prediction values of Crack are 0.99,
0.96, 0.99, and 1.00. The prediction values of snowflake are 0.96, 0.88,
0.94, and 0.95. From the figure, the prediction performance of the
improved RetinaNet for the validation set of surface defects is shown.