基于改进YOLOv5算法的晶圆表面缺陷检测方法

Wafer Surface Defect Detection Method Based on Improved YOLOv5 Algorithm

为了兼顾实时性和准确率,提出了一种基于改进YOLOv5算法的晶圆表面缺陷检测方法。该方法采用了轻量级网络GhostNet作为主干提取网络,以降低模型复杂度并提升检测速度。同时为了提高模型的特征提取能力和检测精度,引入了高效通道注意力机制。此外采用FReLU激活函数取代了原有的SiLU函数,以增强模型对空间的敏感性,提高检测准确性。使用真实的晶圆缺陷数据集对改进模型进行验证。实验结果表明,相比于原始模型,改进YOLOv5网络模型实现了30.02%的参数压缩,同时目标精度达到78.6%,相较于YOLOv5s提升了4.4%,mAP值提高5.5%,检测速度提高1.3 ms。

Wafer surface defect detection holds significant importance in semiconductor chip manufacturing.However,during the inspection process,false detection and missed detection of defects often occur due to the complexity and diversity of wafer surface defect types and manifestations.To balance real time and accuracy requirements,a wafer surface defect detection method based on the improved YOLOv5 algorithm is proposed.This method uses the lightweight network GhostNet as the backbone extraction network to reduce model complexity and improve detection speed.Additionally,an efficient channel attention mechanism is introduced to enhance the model's feature extraction ability and detection accuracy.The original SiLU function is replaced with the FReLU activation function to improve the model's sensitivity to space and detection accuracy.The improved model is validated using a real wafer defect dataset.The experimental results show that the improved YOLOv5 network model achieves 30.02%parameter compression compared with the original model.The target accuracy reaches 78.6%,which is 4.4%higher than YOLOv5s.The mAP value is increased by 5.5%,and the detection speed is increased by 1.3 ms.