Semantic segmentation of viaduct parts

The project main deliverable is a computer software for automatic detection of defects in a concrete structure. The software is available in two versions. One version is written in Python and the second one is based on JavaScript programming language. Both version are provided along with a user-friendly graphical interface. Using web-based application developed in JavaScript the end-users are able to train the proposed architecture of U-net neural network using a stationary computer or a laptop in a setandart web browsers. Additionally, they can apply pre-trained networks on a smartphone or a mobile device to identify damages in real-time. Image processing algorithms are implemented with aid of packages such as Keras and TensorFlow. The user-friendly application can be useful in automatic detection of cracks in concrete structures for people, who are responsible for the condition assessment of the structure.