Low-Resolution Image Classification of Cracked Concrete Surface Using Decision Tree Technique

Abstract

Cracks are essential for assessing the quality of concrete structures since they influence the structure’s safety, application, and durability. Cracks on the concrete surface are one of the earliest signs of structural damage, and detecting the crack is essential for maintenance. The first step in a manual examination is to sketch the crack and note the conditions. A lack of impartiality in quantitative analysis from the manual approach is utterly reliant on the specialist’s knowledge and experience. As an alternative, automated image-based crack detection is suggested. There are many features extraction and classification techniques available for crack detection, including the k-nearest neighbors (KNN), Artificial neural network (ANN), and Decision Tree (DT). This paper aims to detect the building cracks using low-resolution images where KNN, ANN, and DT were trained and evaluated with different images sizes of 50 × 50, 35 × 35, 25 × 25, 10 × 10, and 5 × 5. On the sample images 50 × 50 and 5 × 5, the DT classification approach produced the highest precision values of around 90% to 95%, compared to the other two techniques, KNN and ANN, which provided 76% to 86% and 93 to 88%, respectively. The new findings revealed that KNN, ANN, and DT algorithms give high accuracy with the low-resolution image of 5 × 5 as with the higher resolution image of 50 × 50.