Crack Detection In Concrete Using Artificial Intelligence With Deep Learning

Resumo

Artificial Intelligence (AI) is a field that has been drastically changing not only several areas of
knowledge, but it also brings high expectations regarding the future of professions. While there are projections of
great growth in demand for data scientists, there is the possible threat to unqualified labour, where AI can offer a
low-cost alternative. Deep Learning is a subset of Machine Learning, which is a field dedicated to the study and
development of machines [1], which can be seen as a stage of AI. Also called Deep Neural Network, refers to
Artificial Neural Networks (ANN) with multiple layers. In recent decades, it has been considered one of the most
powerfull tool, and has become very popular in literature as it is able to deal with a great amount of data. Interest
in having deeper hidden layers began recently to overcome the performance of classical methods in different fields,
especially in pattern recognition [1]. Neural networks are used in many areas, such as search algorithms on search
engines, content recommendation algorithms, autonomous cars, speech recognition (audio), natural language
recognition (text) and computer vision (images). The use in image recognition is mainly done with Convolutional
Neural Networks. Thus, the present work intends to apply Convolutional Neural Networks for the detection of
cracks in concrete structures through image processing, especially the obtained with drones. The detection of
cracks by visual inspection can be a very laborious process, depending on the number of cracks and the difficulty
of access, in addition to relying heavily on the subjectivity of the observer. Thus, several methods have been
proposed to automate this process, which consist of image processing techniques. However, the implementation
of these techniques is difficult when there are adverse conditions, such as changes in different textures [2]. It is in
this sense that the use of neural networks brings the expectation of being a method appropriate in relation to the
stability in the detection, even considering variations in the conditions for acquiring images, such as lighting, angle
of acquisition, texture, dimensions, among others. This expectation is mainly due to the ability to automatically
learn the characteristics relevant to the detection of cracks, whereas there are adverse conditions in the learning
data.