Objective  With the development of tunnel engineering and advancement of science and technology, traditional tunnel crack detection methods primarily adopting manual inspection and crack detectors suffer from low efficiency, long operational cycles, and insufficient intelligence, severely restricting the normal construction, operation and maintenance of tunnel projects. To address these issues, a tunnel lining surface crack identification algorithm based on the VGG16 convolutional neural network (hereinafter referred to as the "VGG16 crack identification algorithm") is proposed.

Method  First, illumination correction is performed on the images of the tunnel inner wall to maintain uniform light intensity across image patches. Second, the VGG16 convolutional neural network is employed to predict the crack attributes of the pre-processed image patches. Finally, based on the prediction results, the crack attributes are corrected and optimized to improve identification accuracy. Combined with different crack attributes, precise measurement of crack width is achieved, effectively avoiding the problem of crack misjudgment.

Result & Conclusion  The proposed VGG16 crack identification algorithm demonstrates high accuracy in detecting cracks on tunnel lining surfaces. Field test data validation indicates that the algorithm can identify cracks with a minimum width of 0.2 mm, significantly enhancing the detection efficiency, with an identification time of only 1 s per image. The proposed VGG16 crack identification algorithm can rapidly identify tunnel lining cracks and accurately determine their development trends, providing a scientific basis for tunnel maintenance and offering important technical support for enhancing the safety assessment and O&M assurance capabilities of tunnel engineering.