Corrosion defect segmentation method based on superpixel feature cascade

Abstract

To solve the segmentation problem caused by the small number of feature points and the change of image brightness on the surface of the storage tank, a corrosion defect segmentation method based on the superpixel feature cascade is proposed in this paper. First, the image is segmented to generate superpixels and color and texture features are extracted in the superpixel region and concatenated with domain superpixel features to form superpixel level context features; Then, a plurality of superpixels are labeled according to the pixel range of corrosion defects and the labeling results are obtained; Then, the relationship between superpixels is modeled by the full connection CRF model to optimize the classification results; Finally, the classification result of the input image is refined to generate a segmented image. The effectiveness of the segmentation method is verified and analyzed by designing comparative experiments.

Keywords

• Rust detection
• Superpixel feature cascade
• SVM
• Fully connected CRF