A Deep Learning-based Method for Metal Surface Defect Recognition
Aiming at the problems of various defect types and variable sizes and shapes in the surface defect recognition of metal products,this paper proposes a deep learning model based on multi-scale residual convolutional network.First,ResNet50 is used as a feature encoder to extract feature maps with different resolutions to capture multi-scale information,thereby improving its ability to identify defects of different sizes;Then,a multi-layer perceptron(MLP)is applied to adaptively fuse multi-scale features,where the features such as image textures and boundary information obtained by shallow convolutions and complex semantic information extracted by deep convolutions are used for information interaction and feature refinement to improve the recognition performance of the network.The experimental results show that the proposed algorithm achieves an accuracy rate of 98.06%on the NEU-DET dataset,which has higher recognition accuracy than other models.
deep learningdefect recognitionmulti-scale featuresMulti-Layer Perceptron
NETL
NSTL
万方数据
维普
