基于特征嵌入的小样本涡轮叶片缺陷识别研究

Research on small-sample turbine blade defect identification based on feature embedding

航空发动机涡轮叶片的缺陷,影响发动机可靠性与使用寿命,基于计算机视觉与深度学习技术进行叶片缺陷的自动化检测具有重要现实意义。但是,涡轮叶片图像采集环境的高度非结构化、缺陷形式高差异性,为准确的缺陷识别带来困难。针对上述问题,提出了深度特征嵌入先验网络,其核心通过引入缺陷形状先验的特征嵌入层,准确刻画缺陷的形状特征,提高模型在小样本情况下的分类准确率。实验结果表明,所提方法在小样本叶片缺陷识别问题上取得了优越性能。

Defects in aero-engine turbine blades affect engine reliability and service life,and automated defect detection based on computer vision and deep learning technologies is of practical importance.However,the highly unstructured environment of turbine blade image acquisition and the substantial variation in defect forms pose challenges to accurate defect identification.To address these issues,a deep feature embedding prior network is proposed.The core of this approach involves introducing a feature embedding layer with defect shape prior knowledge to accurately capture the shape characteristics of defects,thereby improving the classification accuracy of the model under small sample conditions.Experimental results demonstrate that the proposed method achieves superior performance in small sample turbine blade defect recognition tasks.