Crack Detection of Concrete Surfaces with A Combination of Feature Extraction and Image-Based Backpropagation Artificial Neural Networks

Abstract

Concrete surface imperfections can signify a structure undergoing severe degradation. It deteriorates when concrete is exposed to elemental reactions such as fire, chemicals, physical damage, and calcium leaching. Due to its structural degradation, concrete deterioration poses a risk to the surrounding environment. Structural buildings can collapse due to severe concrete decline. To prevent concrete cracks early, it is imperative to identify the concrete surface. This requires the development of a technique for detecting the image-based concrete surface. One way to detect concrete surfaces is to create artificial neural networks. The purpose of this study is to combine feature extraction and artificial neural networks to detect cracks in concrete surfaces. The data used is concrete surface image data divided into two classes, namely cracked class and uncracked class. The total data is 600 data points, 300 data points, and 300 data points. The technique used is feature extraction from GLCM and Backpropagation Artificial Neural Network. Test results using epoch five show 95% accuracy, epoch 10 shows 95% results, epoch 100 shows 83% accuracy, and epoch 250 shows 73% results. The test results that have been carried out show a decrease in lower accuracy results when the epoch is determined to be higher. The results of this study epoch that shows the highest accuracy results are epoch 5 with 95% accuracy and epoch 10 with 95% accuracy.

Keywords

• artificial neural network
• backpropagation
• concrete crack detection
• disaster
• feature extraction