A deep learning model based on multi-source feature fusion and defect type recognition method for complex components

The internal defect type recognition of complex components is of great significance to ensure equipment manufacturing quality and safe and reliable operation. Aiming at the problems of poor local feature extraction, lack of considering the empirical characteristics of defects and loss of feature information in the existing deep learning model for defect type recognition, a defect type recognition method based on prior features, global features and ReliefF-Pooling strategy is proposed. The fusion analysis of prior features such as defect geometry, texture and global features of convolutional neural network (CNN) is realized, and the feature optimization method based on ReliefF-Pooling is constructed to optimize the utilization of feature information with different weights. Finally, as an example, the actual radiographic testing of internal defects in complex components in an aerospace enterprise is used to verify the proposed method. The experimental results show that the proposed method can effectively improve the type recognition accuracy of internal defects in complex components.