A Surface Crack Detection System Based on Image Input: the Cnn Approach

Abstract

Any concrete construction that has surface fractures can seriously harm both its surroundings and the people nearby. Cracks that are discovered early on can help stop additional harm. Cracks are found using traditional techniques, which entail visual inspection by humans. The construction sector has its problems and complexities, identical to any other industry. Any pro blems that develop during construction can be found and fixed immediately. Due to time and money limitations, it is challenging to visually detect cracks and other flaws in particularly big constructions. This will cost an enormous amount of money, materials, labor, and time. Additionally, if cracks are found manually, the areas where they can be found are constrained, and human error is a possibility. This research study proposes a novel autonomous system that acts as surface recognition model using CNN with multiple layers, as well as basic preprocessing. To solve these issues, data enhancement has been done in this model to split the obtained data into two broad categories: The crack occurs means positive, and the crack does not exist meaning negative. The suggested surface classification approach achieved a precision of 99.51 % and 99.13%, respectively. Since pa vement or roadway cracks will be automatically identified and analyzed, fewer accidents will occur on roads because of this classification. When compared to manual observation, the suggested system will also be quicker. © 2024 IEEE.