발간년도 : [2023]
| 논문정보 |
|
| 논문명(한글) |
[Vol.18, No.6] Automotive Tire Defect Classification Based on Deep Learning |
|
| 논문투고자 |
Seungchul Kim, Sangmin Suh |
|
| 논문내용 |
Preventing traffic accidents is one of the most important things you can do, as they can be verytraumatic for the driver, passengers, and other drivers. One of the causes of traffic accidents is defectivetyres. Many studies have been conducted to prevent tyre defects, and in this paper, we explore how todetect tyre defects on mobile devices by creating a deep learning model with a small number ofparameters. In this study, we use two classes, "good" and "defective," to distinguish the status of tyres.To balance the dataset, we used 800 images for each class, and 90% of the dataset was used fortraining and 10% for validation and test. To optimize the deep learning model, we used theAdamW(Adaptive moment estimation + Weight decay) optimization function. The hyper-parameters ofthe training model were batch size 32, epoch 300, and learning rate 0.000002. We use a ConvolutionNeural Network(CNN) with ReLU Activation Function. We use a (3×3) and (1×1) Convolution filters totrain our model, and use BatchNormalization layer to avoid biased distribution. We evaluated theperformance of the trained model and found that the F1-Score of the "good" class was 0.863, theF1-Score of the "defective" class was 0.874, and the average F1-Score was 0.869. We also used ROC(Receiver Operating Characteristic) Curve and AUC (Area Under Curve) as other performance evaluationmethods, and the AUC of "good" was 0.94 and that of "defective" was 0.94. |
|
| 첨부논문 |
|
|
|
|
|
자료실 > 논문지 
