MAI Zeitgeist: how to solve an IT case and climb the Aviahackathon podium

On October 21-23, the MAI IT Center, Institute No. 8 "Computer Science and Applied Mathematics" and the WINNO Moscow innovation accelerator held the V International Aviahackathon at MAI new IT site. The event gathered about 800 participants. 166 of them managed to get to the final. The teams solved one of five cases presented by MTS, S7 Technics, Sheremetyevo International Airport, Geoscan and MAI IT Center.

As a result, among the winners there was the Zeitgeist team (MAI), that brought together students of the master's program "Technologies of supercomputer modeling of complex technical Systems": Vsevolod Peshikov, Valeria Samsonova, Ilya Stulov, employees of NIO-101, and Viktor Chistov and Ivan Skvortsov. They presented the best solution of the case from the Geoscan group of companies "Forest taxation: recognition of deciduous and coniferous tree species". We talked to the team and learned about their path to IT and how exactly they managed to become the champions of Aviahackathon.

The captain of the Zeitgeist team Vsevolod Peshikov graduated from Institute No. 1 "Aviation Technology", "Aircraft and helicopter construction" direction. At the 4th year he started working in the research department of the department 101 in the laboratory No. 1 "Mathematical modeling".

–  At NIO-101 I am programming in C#. Since I have been looking for opportunities to get an education in IT for a long time to expand my competencies and deepen my knowledge, I decided to go to the master's program of the IT Center," he says.

Vsevolod participates in the MAI Aviahackathon for the second time, but the rest of the participants are newcomers.

– I decided to participate and was not disappointed. This event corresponds to the sphere of my interests in programming, and in general it is a good way to find new connections and acquaintances," says team member Ilya Stulov.

To solve the case the participants had to develop algorithms for image segmentation, classification of selected segments, analysis of characteristics of selected segments using the available remote sensing data of the Earth – initial aerial photographs, orthophotomaps collected from them, altitude data presented by a digital altitude model, and laser scanning data.

– The task was as follows: you have an aerial photo, an orthophotomap, an altitude map. You need to get a mask for the crowns of trees on an orthophotomap with information about the total number of trees, the area occupied by the crowns, the characteristics of the crowns and the type of trees as a result. The most difficult thing was to start training the neural network, as there were problems with the hardware. We went through dozens of python core crashes due to memory overflow of the GPU where the training took place. In the second most difficult place, I would put the creation of a dataset (processed and structured data array - ed.) as a very time-consuming process: we had to do it three times, which in total took us about 15 manhours," says Vsevolod.

In the team, Vsevolod, as the captain, was engaged in planning, preparing a data set, on which the neural network studied, presenting, sending requests to experts and curator, defending the project at checkpoints and in the finals, preparing reports and helping with algorithms. Valeria was responsible for the presentation, and also helped in the search and information organization of work. Ivan is an expert in artificial intelligence, neural networks, writing code for programs, and designing a code repository. Ilya was responsible for cutting a large orthophotomap into parts and creating data sets for the neural network training. Victor was responsible for training the neural network and displaying the results of its work, writing code and preparing the data set.

As a result, the team managed to compile a dataset, train a neural network, make a script that works with marked-up pictures, an augmentator script and algorithms that calculate the area occupied by tree crowns. The neural network learned to create masks that highlight the crowns of trees in pictures compiled by cutting an orthophotomap.

Programmers from Zeitgeist believe that the fight for the first place was not easy – the rivals were strong enough.

– In the final, one team impressed us - there were even schoolchildren there. They found a thematic library from the University of California and dealt with it. Another team consisted of only two people. They managed to show very interesting intermediate results, which, unfortunately, they did not have time to complete," Vsevolod notes.

The ideas of the winners of Aviahackathon are successfully implemented annually in partner companies. The solution proposed by the Zeitgeist team will also not be left without perspectives. In the near future, the guys will meet with the management of the company to discuss further cooperation.

We asked the guys how to win the hackathon. Valeria Samsonova said that you need to keep positive attitude, be organized despite fatigue, and respond to every difficulty with laughter.

However, in addition to this, you need to be inquisitive and active: develop your skills and competencies not only within the framework of the main direction, but also in related fields. So, for example, Vsevolod was trained at the School of Service and the School of Management of MAI, became a finalist of the Olympiad "Umnik-MIPT-2018", in 2019 took first place in the All-Russian Student Olympiad "Computer Modeling in Aircraft Engineering" named after M.P. Simonov in the nomination "CAD technologies in the Solid Works environment", became finalist of the competition "The Future of aviation-2020" from PJSC "UAC". Valeria, Ivan and Victor graduated from the first higher education in MAI with a red diploma, Valeria also studied at the MAI Service School, where she defended with the team the project of developing a return-exchange fund for the Ansat helicopter. Victor successfully defended his final qualifying work on the topic of recognizing objects in terrain images using convolutional neural networks, and Ivan leads the SQL expertise department at MTS Digital.

– For me personally, the victory confirms that we worked well together as a team and chose the right path for our professional development, — says the captain of the MAI team.

He notes that work and study take up almost all of his free time.

"But I have a wonderful cat and a small ant farm," he says.