I am happy to announce that, together with professor Kyosuke Yamamoto, I was awarded the Wakate Yuugo grant for young multidisciplinary scientists, for the project "Concrete Crack Detection and Identification based on Image Recognition with Machine Learning".
The idea of this project is to create a system that can automatically detect the existence of failures in concrete structures (such as bridges and tunnels) based on the shape and size of cracks seen in photos from these structures.
Imagine this scene: An automated car with a camera on top drives under the tonnels and takes pictures of its interior. These pictures are sent to this system which, based on what it learned from similar data, decided whether any cracks indicate the need for a human engineering to make an on-site assessment of the health of the structure.
For a country such as Japan, where its infrastructure is under constant assault from the forces of Nature, this sort of automated diagnostic is essential.
Some hard parts of this research include separating the real cracks from dirt and imperfections in the image, taking into account different light conditions from pictures of the same concrete tunnel, and learning to differentiate dangerous cracks from harmless ones borne out of superficial wear and tear.
The grant money will go a good way into helping us hire people to collect and pre-process training data, and into buying extra computers for image processing and algorithmic fine tuning.