Deterioration Detection in Historical Buildings with Different Materials Based on Novel Deep Learning Methods with Focusing on Isfahan Historical Bridges

Historical bridges comprise part of any society's history, culture, and identity and reveal the manufacturing technology of hydraulic structures at their time. Nevertheless, these structures deteriorate because of their materials, the passage of time, and natural factors. Because of drought in the past two decades, historic bridges in Isfahan have faced consecutive wet-dry cycles, resulting in further defects in bridges. Moreover, stone materials in the bases and brick materials in the bodies of bridges have made detecting defects more complex, requiring experts for each material. Additionally, insufficient attention to these defects or human errors in their proper detection can affect their structural integrity. This article has utilized deep learning methods to detect defects in these structures with different materials. To achieve initial data, the authors took 8331 images of bridges in Isfahan. Then, the defects (cracking, flaking, erosion, salt efflorescence, and no defect) were labeled based on the materials (brick and stone). Overall, seven different classes were defined for network training. After investigating various models of deep networks, the Inception-ResNet-v2 model was selected as the optimal model. We used this model to achieve the accuracy, precision, and recall criteria of 96.58, 96.96, and 96.24%, respectively.