Application of MobileNetV2 to waste classification

Area of Science:

• Computer Science
• Environmental Science
• Artificial Intelligence

Background:

• Domestic waste management faces challenges with manual sorting, leading to inefficiencies.
• Existing waste separation devices have limited impact on overall domestic waste management.
• Automated classification is crucial for improving recycling rates and reducing landfill burden.

Purpose of the Study:

• To apply deep learning for automated domestic waste classification.
• To develop a model capable of distinguishing between recyclable, kitchen, hazardous, and other waste.
• To evaluate the performance of a MobileNetV2-based model for waste sorting.

Main Methods:

• A deep learning approach utilizing the MobileNetV2 neural network architecture.
• Training a classification model on a dataset of domestic waste images.
• Comparing the MobileNetV2 model's performance against a Convolutional Neural Network (CNN) model.

Main Results:

• The MobileNetV2 model achieved an absolute accuracy of 82.92% in classifying domestic waste.
• The MobileNetV2 model demonstrated a 15.42% higher classification accuracy compared to the CNN model.
• The trained model is lightweight, suitable for mobile applications.

Conclusions:

• Deep learning, specifically MobileNetV2, offers an effective solution for automated domestic waste classification.
• The developed model significantly improves accuracy and efficiency over traditional methods.
• The model's mobile compatibility facilitates broader application in real-world waste management scenarios.