Automatic Defect Recognition Method of Aluminium Profile Surface Defects

Automatic defect detection has important implications to intelligent manufacturing which could be used for the precise quality control of different products. However, the diverse aluminium profile surface defects present the characteristics of micro defects and different sizes. Conventional handcrafted-based methods and machine learning-based methods have limited feature expression ability which cause relatively poor detection performance. Recently, with the stronger feature extraction ability, deep learning has got wide applications on defect detection and recognition. Due to the loss information caused by pooling operations, it still exists a certain drawbacks on multi-scale object detection. To address this issue, with the residual neural network (ResNet), a new deep defect recognition network is proposed in this paper for aluminium profile surface defects to construct an end-to-end defect detection scheme. An attention fusion model is proposed to improve the detection precision on multi-scale defects. Experiments show that the proposed defect detection method shows a better detection performance compared with other advanced detection models.