A Domestic Trash Detection Model Based on Improved YOLOX.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Domestic trash detection is crucial for smart cities, but current algorithms struggle with accuracy, false positives, and speed.
• Existing methods face challenges due to the complexity and variability of urban trash scenarios.

Purpose of the Study:

• To propose an improved deep learning model, i-YOLOX, for accurate and efficient domestic trash detection.
• To enhance feature extraction and model convergence for better trash classification and regression.

Main Methods:

• Developed a new trash image dataset and incorporated the involution operator for long-distance feature relationships.
• Integrated the convolutional block attention module (CBAM) to improve discrimination of similar trash features.
• Designed an involution residual head structure to mitigate gradient disappearance and accelerate model convergence.

Main Results:

• The i-YOLOX model achieved a 1.47% increase in mean average precision (mAP) compared to the YOLOX-S baseline.
• Reduced the number of parameters by 23.3% and improved Frames Per Second (FPS) by 40.4%.
• Demonstrated accurate trash recognition in natural scenes, validating its generalization performance.

Conclusions:

• The i-YOLOX model offers a significant advancement in domestic trash detection technology.
• This research provides a valuable reference for future developments in smart city waste management systems.