YOLOv8-seg-CP: a lightweight instance segmentation algorithm for chip pad based on improved YOLOv8-seg model

Area of Science:

• Computer Vision
• Machine Learning
• Semiconductor Manufacturing

Background:

• Accurate chip pad detection and segmentation are crucial for automated chip alignment and position correction.
• Existing methods face challenges with small target detection, segmentation accuracy, and model efficiency.

Purpose of the Study:

• To develop a lightweight and efficient chip pad instance segmentation algorithm.
• To enhance detection and segmentation accuracy for small chip pad targets.
• To reduce model size, computational load, and inference time.

Main Methods:

• Integration of StarNet into the YOLOv8-seg backbone for improved feature capture.
• Development of the C2f-Star module in the neck network for enhanced small target feature extraction.
• Introduction of a lightweight shared convolution segmentation head (LSCSH) and a CGCAFusion module for improved segmentation and feature fusion.

Main Results:

• Achieved 89.8% detection and segmentation accuracy on the chip pad dataset.
• Reduced model size by 45.6%, parameters by 50%, and FLOPs by 31.7% compared to the baseline.
• Improved Frames Per Second (FPS) by 25.8% and reduced inference time by 0.18ms.

Conclusions:

• The proposed YOLOv8-seg-CP algorithm offers superior performance in chip pad detection and segmentation.
• The lightweight design and enhanced feature extraction contribute to significant improvements in efficiency and accuracy.
• This approach demonstrates high industrial application value for automated chip testing and sorting.