Automated Crack and Damage Identification in Premises using Aerial Images based on Machine Learning Techniques

The impartiality and reliability of evaluation, as well as the high time and expense demands, make it impossible to conduct a manual examination of infrastructure issues such as building fractures. For airborne images of damage, use unmanned aerial vehicles. Artificial intelligence and machine learning methods may help overcome the limits of many computer vision-based approaches to crack detection. But these hybrid approaches have their own limitations that can be solved. Images with damage may be more accurately detected using modified convolutional neural networks (MCNNs), which are less affected by picture noise. For fracture identification and damage assessment in civil infrastructures, a Modified Deep CNN Model (MDCNN) has been deployed. The 16-layer convolutional architecture and the Support Vector Machine are used in this design. The last layer of the CNN networks is replaced with SVM. Rather of relying on a single layer, we suggest a multi-layered network instead. Their abilities in identifying objects and putting them into categories are quite reliable. A further great benefit of MDCNNs is their ability to share the burden. When compared to a standard neural network, proposed method use significantly less processing power.