The article studiesthe possibility of using the machine vision while ice navigation in order to automate the process of detecting "tracks" in ice fields. The technology of image processing for detecting ice fractures’ boundaries with help of MobileNet convolutional neural network architecture and deep learning is presented. A method of calculating the theoretical line of the ice "track" using linear regression is described. A software product has been created and it is capable to perform ice "tracks" automated detection and to calculate their deviation from the true course of the vessel. The limitations and disadvantages of the program, as well as its practical importance in the prospects of automatic navigational systems, are described. Program shows an example of implementation on Arc7 TNG carrier.
machine vision, arctic navigation, ice navigation, automation of navigation, "tracks" in ice, convolutional neural network, deep learning, neural networks, linear regression, TNG vessel, Arc7 TNG
1. Giniyatova D. Neyronnye seti na python. - Kazan': Kazanskiy Federal'nyy Universitet, 2017. - 7 s. Rezhim dostupa: https://kpfu.ru/portal/docs/F 1458204831/Neiroimve.seti.na.Python.pdf
2. Gonsales R., Vuds R. Cifrovaya obrabotka izobra zheniy // Tehnosfera - 2005. - 1007 s
3. Il'in IV, Gudkov K.V. Analiz programmnyh sredstv dlya glubinnogo obucheniya iskusstvennyh neyronnyh setey,- Penza: Penzenskiy gosudarstvennyy tehnologicheskiy universitet, 2018.-7s
4. Sozykin A. V. Obzor metodov obucheniya glubokih neyronnyh setey // Vestnik Yuzhno-Ural'skogo Gos. Universiteta-2017-Tom 6-№ 3 -31 s
5. Tul'chinskiy V.V, Tul'chinskiy V.I. Rassmotre nie vozmozhnosti primeneniya elementov mashinnogo zreniya dlya avtomatizacii ledovoy provodki na primere SPG gazovozov tipa Ags7. // Ekspluataciya morskogo transporta,- 2019.- №3- S. 67-76
6. Shashkov V.B. Prikladnoy Regressionnyy analiz,- Orenburg: Orenburgskiy gosudarstvennyy universitet, 2003.-363 s
7. Howard A.G., ZhuBoChen M., Kalenichenko D., Wang W., Weyand T., Andreetto M., Hartwig A. MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications.Google Inc., 2017. -9 p
