Image-Based Automatic Crack Detection Method of Gantry Crane Track

Abstract

Aiming at the technical problems of intelligent identification of cracks on gantry crane track, a target detection method was proposed based on machine vision and deep neural network. YOLOv5s is used as the surface defect detection model, and the trained YOLOv5s model is used to recognize the crack of the gantry crane track. Because of the lack of training sets, the affine transformation is used to increase the sample size of the training set, and the data set is trained to get preliminary weight for the transfer learning method. Also, the Gaussian smoothing image method is adopted to improve the model's accuracy. The test results show that the track crack detection model developed based on YOLOv5s is effective. The test accuracy P, recall rate R, and mAP reach 0.995,0.993, and 0.996, respectively. The developed method is helpful for the construction personnel to evaluate the health condition of gantry crane track crack accurately.