Moisture Detector in Concrete using Convolutional Neural Networks

Resumo

Concrete undergoes various indicators of the deterioration processes, such as pH changes, compressive strength reduction, and microbe growth, all associated with moisture. As a result, wet concrete can lose its adhesive properties and compressive strengths, leading to failure. To address this issue, this study presents a classifier that uses convolutional neural networks (CNNs) with a custom and low-complexity architecture to detect surficial signs of wet concrete visible to the human naked eye. The sample used in the study was built upon scraped data from the open database, as well as authorial photographs. The images of concrete surfaces were divided into two classes: with and without moisture. The results indicate that the classifier can effectively classify images with visible moisture that the human eye can detect, with good performance.