Su Dongjie
ABSTRACT
This study aimed to design and develop a waste sorting system suitable for individual household use, which can help people differentiate waste categories, carry out waste sorting and disposal, and encourage the habit of proper waste management. The system was developed using convolutional neural networks, with a fine-tuned ResNet50 pre-trained model applied to adapt to the garbage classification task. Through transfer learning, the model converged quickly and achieved high classification accuracy. The graphical user interface was implemented using the PyQt5 framework, providing users with a friendly and intuitive platform for waste classification operations. The findings showed that the integration of deep learning, image processing, and graphical user interface technologies resulted in an efficient and accurate garbage classification system. The system was able to process uploaded waste images, classify them effectively, and provide classification probabilities. Users could also submit misclassified images to further optimize the model. Overall, the system demonstrated compliance with ISO/IEC software quality standards to a very large extent, highlighting its potential as a practical tool to support ecological initiatives and promote sustainable waste management practices.
Keywords: Deep learning, image processing, ResNet50, waste classification, waste management system
https://doi.org/10.57180/pajq8986