Method for white-light interference fringe detection based on improved YOLOv5

In white-light interferometry (WLI) for surface topography measurement, the width of interference fringes directly reflects the surface information of the sample under test. Operators rely on this to adjust the sample posture for high-precision measurements. Consequently, there is an urgent need to develop a technique capable of rapidly locating the central region of white-light interference fringes. This study proposes a fast detection method based on an improved YOLOv5 model. Leveraging YOLOv5's fast inference capability, the method efficiently locates the central fringe region. Enhancements are introduced through the incorporation of a coordinate attention (CA) module and a weighted bidirectional feature pyramid network (BiFPN), achieving superior feature fusion and strengthening the model's ability to detect fringes of varying sizes. Compared to the original YOLOv5 model, the mean average precision (mAP) of the improved model increased from 72.1% to 82.8%. The experimental results indicate that the method can quickly and accurately detect white-light interference fringes, showing strong transferability and generalization ability in different instruments and complex scenarios, thus providing the possibility for real-time detection.