HGCS-Det: A Deep Learning-Based Solution for Localizing and Recognizing Household Garbage in Complex Scenarios.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Environmental Technology

Background:

• Deep learning offers potential for intelligent garbage detection and classification.
• Complex environments and irregular garbage features pose significant challenges to current detection methods.
• Existing models often compromise between precision and real-time performance in complex garbage detection scenarios.

Purpose of the Study:

• To propose a novel deep learning model, HGCS-Det, for robust garbage detection in challenging environments.
• To enhance the precision and real-time capabilities of garbage detection systems.
• To provide a practical solution for real-world garbage classification and management.

Main Methods:

• Developed HGCS-Det based on YOLOv8, incorporating a normalization attention module to reduce noise interference.
• Integrated an Attention Feature Fusion module to optimize channel attention weights.
• Employed an Instance Boundary Reinforcement module for fine-grained feature extraction and a Slide Loss function to improve hard sample recognition.

Main Results:

• HGCS-Det achieved 93.6% mean average precision (mAP) and 86 frames per second (FPS) on the HGI30 dataset.
• The model demonstrated a 3.33% higher mAP compared to YOLOv12 with a minimal parameter increase (3.02M).
• Outperformed state-of-the-art methods in both detection efficiency and applicability.

Conclusions:

• HGCS-Det offers a lightweight yet accurate solution for garbage detection in complex scenarios.
• The model's real-time performance and enhanced accuracy make it suitable for embedded systems and practical waste management.
• This research provides a valuable technical reference for engineering applications in garbage classification systems.