Optimized car parts detection with advanced feature fusion and attention modules

Area of Science:

• Computer Vision
• Artificial Intelligence
• Automotive Engineering

Background:

• Accurate car part detection is crucial for intelligent transportation systems, automated vehicle inspection, and maintenance.
• Challenges include varying object scales, background clutter, and occlusions, hindering reliable real-time detection.

Purpose of the Study:

• To present an enhanced YOLO-based architecture for fine-grained car-part detection.
• To improve accuracy, robustness, and efficiency in real-time car part identification.

Main Methods:

• Developed an enhanced YOLO architecture integrating task-specific feature refinement and improved supervision.
• Incorporated modified C2fCIB block, improved PSA module, SPPF layer, SCDown module, and a Dual Assignment Head.
• Utilized a car-parts dataset for experimental validation.

Main Results:

• Achieved 63.3% precision, 81.6% recall, 73.7% mAP, and 111 FPS inference speed.
• Outperformed baseline detectors like Faster R-CNN, SSD, YOLOv4, YOLOv5, YOLOv7, and YOLOv8.
• Demonstrated enhanced small-part sensitivity, localization robustness, and recall.

Conclusions:

• The proposed architecture offers an effective balance of accuracy and efficiency for car part detection.
• The model is suitable for real-world automotive inspection and intelligent vehicle applications.