Air Tightness Detection of Filters Based on YOLOv8-Bubble

This paper proposes a filter quality detection solution based on the combination of YOLOv8 and the immersion bubble method. Building upon YOLOv8, the solution introduces an Efficient Multi-scale Attention mechanism into the backbone network and replaces the original Conv and C2f modules with GSConv and GSConv-C2f modules, respectively. Additionally, the loss function is optimized, significantly enhancing the model's ability to detect small targets, such as tiny bubbles. Experimental results demonstrate that the improved YOLOv8-Bubble algorithm achieves a mean Average Precision of 94%, representing a 6.8% improvement over the original YOLOv8s network. Furthermore, compared to other YOLOv8s-based improved algorithms, YOLOv8-Bubble exhibits superior performance in both detection accuracy and speed, meeting the dual requirements of real-time performance and accuracy for industrial filter inspection.