Integrating Information in Visual and Thermal Image for Crack and Moisture Detection in Concrete Structures through CNNs

The life assessment of buildings is crucial for the stability of the structure and the safety of its inhabitants. Due to environmental factors and ageing, structural deterioration takes place, resulting in fractures and seepage. Therefore, there is a need for inspection tools, preferably non-invasive in nature. Recently the inexpensive cost of the necessary experimental equipment and the general availability of image processing tools have led to its extensive use for damage identification. In this research, an efficient parallel deep-learning model is presented in order to identify cracks in buildings and categorize moisture present in each image. Information from both optical and thermal images has been incorporated into the model. As a first stage of processing, the morphological operation was performed on the thermal images for easier classification of moisture. For moisture classification, the model was able to get an accuracy of 83.33%. The R-CNN network, which was utilized for the detection of cracks, possessed a greater average precision value of 72.27%, along with an intersection over union (IoU) value of 0.5.