Plastic bottle defect detection based on convolutional neural network with MobileNetV2 architecture

Solid waste generated from human activities can be categorized into two main types: organic and inorganic waste. Plastic is one type of inorganic waste that can be recycled to reduce the negative impact of plastic waste and provide economic value. However, manual sorting of plastic bottles is time-consuming and requires a more efficient system to distinguish between recyclable and non-recyclable. This research proposes a plastic bottle defect detection model using a Convolutional Neural Network (CNN) based on the MobileNetV2 architecture with transfer learning. The model was trained using data augmentation, systematic hyperparameter tuning, and fine-tuning with dropout regularization, and evaluated using accuracy, precision, recall, and F1-score metrics. The proposed model was trained and evaluated using a dataset of 1074 plastic bottle images consisting of recyclable and non-recyclable bottle classes. The model used is MobileNetV2 with hyperparameter tuning to achieve optimal performance. The results showed that the best combination of hyperparameters was the Adam optimizer with a learning rate of 0.001 and 20 epochs, achieving an accuracy, precision, recall, and F1-score of 0.98 each . The applied method is effective in classifying images of recyclable and non-recyclable bottles, which provides an experimental contribution toward the development of an artificial intelligence-based waste sorting system and demonstrates its potential for real-world deployment, as well as opening up opportunities for further research in model optimization.