军事知识服务平台
基于改进自注意力机制的模具缺陷识别方法
Mold defect recognition method based on improved self attention mechanism
摘要 为提高模具缺陷识别的精度,提出一种基于改进自注意力机制的模具缺陷识别方法.首先以自注意力机制为基础,通过引入稀疏自注意力模块来降低自注意力机制的空间复杂度,并引入(Kolmogorov-Arnold network,KAN)模块来提高自注意力机制的可解释性,提出改进自注意力机制;然后采用改进自注意力机制对不同类型的模具缺陷进行识别,并开展仿真实验.结果表明,本方法对缺口、磨损、裂纹、气孔等不同类型的模具缺陷识别的平均准确率、召回率、F1值分别为97.07%,97.09%,96.45%,浮点运算数为2.845×104,参数量为2.230×104.相较于其他方法,本方法可更快速、准确识别不同类型的模具缺陷.由此得出,本方法可提高模具缺陷识别的精度.
Abstract:
To improve the accuracy of mold defect identification,a mold defect identification method based on an improved self-attention mechanism is proposed.First,building upon the self-attention mechanism,a sparse self-attention module is introduced to reduce its spatial complexity,and a Kolmogorov-Arnold Network(KAN)module is incorporated to enhance the interpretability of the self-attention mechanism,resulting in the improved self-attention mechanism.Subsequently,the improved self-attention mechanism is employed to identify various types of mold defects,and simulation experiments are conducted.The results demonstrate that for different types of mold defects such as notches,wear,cracks,and pores,the average accuracy,recall rate,and F1 score of this method are 97.07%,97.09%,and 96.45%,respectively,with a floating-point operation count of 2.845×104 and a parameter count of 2.230×104.Compared to other methods,this approach enables faster and more accurate identification of various mold defects.Therefore,it is concluded that this method can enhance the precision of mold defect identification.
刊 名 模具技术 ISTIC
年,卷(期): 2026, (4)
分类号: TP 391
机标分类号: TP391.41 TN911.73 TP273
在线出版日期: 2026年10月19日
基金项目: Y0LOv5;23JK0333:陕西省教育厅专项科研计划项目,