HYFF-CB: Hybrid Feature Fusion Visual Model for Cargo Boxes

Area of Science:

• Computer Vision
• Robotics
• Artificial Intelligence

Background:

• Accurate real-time box detection is critical for automated loading and unloading systems.
• Existing convolutional neural network models struggle with size inconsistency and occlusion in complex truck environments, limiting detection accuracy.
• Current methods fail to meet the stringent requirements of actual production for automated systems.

Purpose of the Study:

• To propose a novel and effective box detection method for automated loading and unloading systems.
• To address the limitations of existing models in handling complex scenarios, including size variations and occlusions.
• To enhance the performance and reliability of automated loading and unloading systems through improved box detection.

Main Methods:

• Development of the HYFF-CB model, integrating a location attention mechanism, fusion-enhanced pyramid structure, and synergistic weighted loss system.
• Acquisition of real-time truck images using an industrial camera.
• Rigorous testing and comparison of the HYFF-CB model against existing box detection models.

Main Results:

• The HYFF-CB model demonstrated significant advantages in detection rate compared to other existing models.
• The model accurately detects stacking locations and quantities of boxes within trucks.
• Detection performance and effect fully meet the application requirements of automatic loading and unloading systems.

Conclusions:

• The HYFF-CB model offers a superior solution for real-time box detection in automated loading and unloading systems.
• The model exhibits excellent adaptability to complex and changing scenarios, overcoming limitations of previous approaches.
• Implementation of HYFF-CB can significantly improve the performance and reliability of automated logistics operations.