Structural damage-causing concrete cracking detection based on a deep-learning method

• Developing deep convolutional neural networks (DCNNs) for image recognition. • Building an intelligent crack identification model using crack feature localisation. • Proposed a novel crack extraction approach combining DCNNs and local threshold segmentation. Cracking is one of the common manifestations of damage in concrete structures. Crack detection is currently done using typical computer vision methods. But they are still not clever enough and have limited precision. Deep convolutional neural networks (DCNNs) are sophisticated models that recently improved visual task performance. This article proposes a hybrid technique based on CNN and digital image processing to identify fractures in photos. Using transfer learning, an AlexNet-based CNN is effectively trained on a tiny dataset and achieves 98.26% accuracy on fresh test data. Then the segments are cracked using a threshold-based technique. In the end, the crack mask is made up of all segments' shows. With transfer learning, the quantity of data and expenses necessary are lowered while retaining accuracy. Precise crack cover detection aids in automated crack measuring and fracture type labeling.