Short description: In the master's thesis, an LSTM autoencoder model is to be developed for the displacement of linear objects using Python and Tensorflow.
Keywords: LSTM autoencoder, automated line displacement, cost function
Deep learning methods have great potential for solving complex generalisation problems, including the displacement of isolated linear objects, such as embankments, or linear networks such as street or river networks. Yu and Chen (2022) showed how LSTM autoencoders can be used to simplify linear objects. Extending this approach, initial experiments by Toth (2023) with isolated lines indicated that displacement problems can also be addressed by extending the cost function of the LSTM autoencoder. The aim of the master thesis is to develop further an generalisation workflow based on LSTM autoencoder for the displacement of isolated linear objects and linear networks. The evaluation can be realised by comparing with existing optimisation approaches such as least squares adjustment, which has been used to displace embankments (Touya et al., 2021).
Literature/references:
Toth, J. (2023). Line generalisation using deep learning. Master thesis, TU Dresden
Touya, G. and Lokhat, I. (2022). ReBankment: displacing embankment lines from roads and rivers with a least squares adjustment, International Journal of Cartography, 8:1, 37-53, https://doi.org/10.1080/23729333.2021.1972787
Yu, W. and Chen, Y. (2022). Data-driven polyline simplification using a stacked autoencoder-based deep neural network, Transactions in GIS, 26, 2302–2325, https://doi.org/10.1111/tgis.12965
Yan, X. and Yang, M. (2022). A Comparative Study of Various Deep Learning Approaches to Shape Encoding of Planar Geospatial Objects. ISPRS Int. J. Geo-Inf., 11, 527 https://doi.org/10.3390/ijgi11100527
