Identification of wheel surface cracks under high noise environment based on texture segmentation and combination descriptor

The wheel is easy to produce defects after a long time of use. Train maintenance technicians must overhaul and maintain train wheels on time and regularly to ensure the safe operation of trains. The existing manual detection has disadvantages of low efficiency and strong subjectivity. Compared with other defect forms, the identification accuracy of wheel surface crack is low due to the high noise and weak features, this paper proposes a feature enhancement and identification method for wheel surface crack based on texture segmentation and combination descriptor. Firstly, a morphology-based texture segmentation method is proposed to separate cracks and noise. Then, the crack characteristics are enhanced by filtering. Finally, a crack diagnosis method based on combination descriptor is proposed, which is evaluated from two dimensions of crack morphology and distribution characteristics. In order to verify the effectiveness of the proposed method, the images of the wheel collected from the freight car workshop were used for verification, and the detection rate reached 93.78%. The experimental results show that the proposed method is more effective in the case of weak crack features and surface dirt, and can better adapt to the identification of surface cracks on wheel in complex engineering applications.