Defect Detection of Airbag Sealing Tape Based On MMR-YOLOv5

This study presents a Multi-Level and Multi-Scale Feature Reconstruction (MMR) module specifically designed for defect detection in airbag sealing tapes. The MMR module enhances contextual interaction across various feature scales and hierarchies, effectively integrating multi-level information to produce richer and more expressive feature representations. Additionally, an optimized decoupled detection head based on depthwise separable convolution is introduced, significantly reducing the model's parameter count while improving computational efficiency. Experiments conducted on both the GC10-DET and Airbag Sealing Tape datasets demonstrate that MMR-YOLO is well-suited for broader applications in defect detection and quality control. MMR-YOLOv5 achieves a 0.39 increase in average precision (AP) compared to the original YOLOv5 and outperforms YOLOv8 in detection accuracy, while using 3.34 million fewer parameters and requiring 11.5 GFLOPs less computation. Furthermore, comparisons with two other lightweight models highlight the superior performance of MMR-YOLOv5, which significantly improves detection accuracy without increasing the model size, achieving an optimal balance between precision and efficiency. This advancement greatly enhances the effectiveness and practicality of defect inspection for airbag sealing tapes, and its successful deployment on edge computing devices further confirms its real-world applicability and robustness.