Crane Rail Bolt Image Defect Recognition Based on Feature Fusion

For the problem of insufficient feature extraction in crane rail bolt image defect recognition method, this paper proposes a feature fusion-based crane rail bolt image defect recognition method. Firstly, the LBP local texture features and Resnet50 deep-level features of the image are extracted respectively; then the two features are fused and optimized using principal component analysis; finally, the optimized feature vectors are inputted into a support vector machine to realize the defect recognition of the crane rail bolt image. The experimental results demonstrate that compared with the typical image single-feature and multi-feature fusion methods, the overall classification accuracy of this paper's method can be improved to 96.93 %, with good robustness.