Kaminskyi, M Nazar-Mykola (2019) Deep learning for predicting ship motion from images PRE - Research Project, ENSTA.
This is the latest version of this item.
Today, artificial intelligence penetrates into all areas of human activity. De- veloped methods and recent advances in machine learning show good results and will gradually replace other ”traditional” approaches. AI allows to automate pro- cesses thereby improving efficiency and accuracy of tasks in co-operation with people. I worked on the problem of predicting ship motion from images of the sea surface. First of all, using 3D graphics generator Blender, I created a dataset, which simulates the ship’s movement through sea waves. This software also gives infor- mation about parameters of the boat (pitch and roll in our case). Data streams were structured in sequences and normalized for further processing. Then 9 dif- ferent neural networks were developed and tested. As I worked with time-ordered image sequences, I decided to use convolutional neural networks (CNN) for pro- cessing images and long short-term memory (LSTM) networks for time series processing. Finally, I run the hyperband algorithm to find the best model hyper- parameters and then all experiment results were analyzed. Also, some possible improvements are suggested.
|Item Type:||Thesis (PRE - Research Project)|
|Subjects:||Information and Communication Sciences and Technologies|
|Deposited By:||Nazar-Mykola KAMINSKYI|
|Deposited On:||09 juin 2021 16:54|
|Dernière modification:||09 juin 2021 16:54|
Available Versions of this Item
- Deep learning for predicting ship motion from images (deposited 09 juin 2021 16:54) [Currently Displayed]
Repository Staff Only: item control page