| Title |
Design of PET Bottle Label Detection System using Teachable Machine |
| Authors |
정경권(Kyung Kwon Jung) ; 김용중(YongJoong Kim) ; 이태원(Taewon Lee) |
| Keywords |
Teachable machine; PET bottle; Label detection; Image classification; Deep learning |
| Abstract |
This paper presents a low-cost recycling support system that combines a deep learning classifier based on Teachable Machine with a control module to determine the presence of labels on transparent PET bottles in real time, physically ejecting labeled bottles to encourage users to remove them. The proposed system captures images using a webcam mounted above the input chute, performs inference using an image classification model trained via Teachable Machine configured with five classes, and transmits the predicted results (Class IDs) to an Arduino via serial communication to control two servomotor-driven gates. The dataset was constructed with 200 images per class and validated by dividing the data into an 85% training set and a 15% testing set. Evaluation resulted in zero misclassifications in the confusion matrix, confirming a 100% classification accuracy. Furthermore, in actual device operation experiments conducting 20 trials per class, all inputs (20/20) were accurately collected or returned, thereby verifying the stable and reliable operation at the system level. |