Wheel Tread Anomaly Detection Based on Attentional Reverse Knowledge Distillation

Table 1. Comparison of results of reverse knowledge distillation model before and after anomaly detection dataset segmentation
View table
Table 1. Comparison of results of reverse knowledge distillation model before and after anomaly detection dataset segmentation
Model Segmentation AUC /% P /% R /% Acc /%
RD Before 66.7 55.5 97.9 58.4
After 93.6 82.1 94.0 86.3

Table 2. Comparison of ablation results
View table
Table 2. Comparison of ablation results
Method AUC /% P /% R /% Acc /%
RD 93.6 82.1 94.0 86.3
RD+Attention（our） 93.8 82.3 95.4 87.0

Table 3. Recall of different types of defects detected before and after model improvement
View table
Table 3. Recall of different types of defects detected before and after model improvement
Method Recall of peel /% Recall of scratch /% Recall of injury /%
RD 97.5 99.0 85.1
RD+Attention（our） 98.0 99.4 88.2

Table 4. Comparison of experimental results of different models
View table
Table 4. Comparison of experimental results of different models
Model AUC /% P /% R /% Acc /%
GANomaly 57.8 51.9 100.0 52.2
MKD 84.3 73.0 88.9 77.3
Our 93.8 82.3 95.4 87.0

Table 5. Recall of different types of defects detected on different models
View table
Table 5. Recall of different types of defects detected on different models
Model Recall of peel /% Recall of scratch /% Recall of injury /%
GANomaly 100 100 100
MKD 95.6 91.7 82.1
Our 98.0 99.4 88.2

Tools

Get Citation

Copy Citation Text

Rongrong Qin, xiaorong Gao, Lin Luo, Jinlong Li. Wheel Tread Anomaly Detection Based on Attentional Reverse Knowledge Distillation[J]. Laser & Optoelectronics Progress, 2023, 60(24): 2415002

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites