Studying Cracks in Infrastructures Using Amalgamation of AlexNet and Moth Flame Optimizer

Crack detection is a vital task to maintain functionality, safety and operation of civil engineering infrastructures. As such, this research paper introduces a computer vision-based model to circumvent the limitations of visual inspection assessment. This model is conceptualized on two main sub-components. In the first sub-component AlexNet deep learning architecture is exploited to extract the global contextual features of the crack images. The second sub-component involves a hybridization of K-nearest neighbors (KNN) and moth flame optimizer (MFO) to learn the features of crack images for recognizing cracks. In this regard, MFO is deployed to amplify the search capabilities of KNN through fine-tuning its structure. It is imparted that the developed model managed to outperform back propagation artificial neural network and Naïve Bayes significantly providing an improvement in the prediction accuracies by 110% and 44%, respectively.