Enhanced ResNet-50 for garbage classification: Feature fusion and depth-separable convolutions

Area of Science:

• Computer Science
• Environmental Science
• Artificial Intelligence

Background:

• Increasing living standards lead to a rapid rise in household garbage generation.
• Effective garbage classification is crucial for resource recycling, environmental protection, and sustainable development.
• Existing deep learning models for waste image classification face challenges with accuracy, robustness, and speed due to high parameter counts.

Purpose of the Study:

• To develop an improved deep learning model for accurate and efficient garbage image classification.
• To enhance model performance by addressing issues of parameter count, computational efficiency, and class imbalance.

Main Methods:

• Proposed a novel garbage image classification model based on the ResNet-50 architecture.
• Introduced a redundancy-weighted feature fusion module to leverage valuable features and reduce parameters.
• Replaced standard convolutions with depth-separable convolutions in ResNet-50 for improved computational efficiency.
• Incorporated a weighting factor into Focal Loss to mitigate class imbalance issues.

Main Results:

• The proposed model achieved a classification accuracy of 94.13% on the TrashNet dataset.
• Demonstrated a significant reduction in model parameters and an improvement in detection speed compared to existing models.
• Effectively handled class imbalance, enhancing overall model robustness.

Conclusions:

• The developed model offers a practical and effective solution for waste image classification.
• The optimizations significantly improve upon existing deep learning approaches for garbage classification.
• The model shows strong potential for real-world applications in waste management and recycling initiatives.