Development of Computational Algorithm for Concrete Leakage Detection Using Deep Convolutional Network

Abstract

Traditional methods for detecting leaks in concrete structures are very expensive, time- consuming, and require human intervention. Since leakage inside a concrete building cannot be seen with the naked eye, another method for detecting leakage is to use a thermal camera. The purpose of this project is to evaluate the use of a convolutional neural network (CNN) to classify leakage and no leakage areas in concrete buildings. The model is built using the Phyton programming language. The model trained for two to four hidden layers that will help in increasing the data accuracy which consists of convolutional layers, pooling layers and the final layer before metric model perform are fully-connected layers. The best accuracy that can be achieved by generating this metric model is more than 95%. At the end of the study, the model performance will classify the 60 collected images as leak or no leak images.