Geoinformatics and Deep Learning methods: A transformation of disciplines

Geoinformatics is an interdisciplinary field that deals with acquiring, processing, analysing and visualising spatial information. Within this discipline, geographic information systems (GIS) are used but can be supplemented or replaced by programming languages such as Python with geo-programming libraries. Spatial data, also known as raster and vector data, contain information about the physical and anthropogenic features of the Earth’s geosystem. It includes, but is not limited to, the topography, vegetation, river systems, and morphology of the Earth, including human impact through the construction and destruction of our landscape. Overarching geoinformatics has applications in many fields, including aerospace, agriculture, urban planning, tourism, disaster management, research, and other sectors.


In particular, the increasing development of machine learning techniques is also influencing geoinformatics, for example, through computer vision techniques, i.e. the ability of algorithms to analyse image data and recognise and classify information contained therein. Such DL methods can be used, for example, to automatically segment landscape classes within a satellite image chronology and thus obtain more in-depth information about the natural changes of our planet Earth.

Architectures used in satellite image segmentation include U-Net (Ronneberger et al., 2015) and ResNet (He et al., 2015). After a long and balanced training, accuracy above 96% in the final test procedure (Tang et al. 2022) is not uncommon with a “convolutional neural network”. Thus, the accuracy of human mapping can be surpassed, and the duration can be significantly reduced.


Artificial neural networks (Ai) will not only replace many a monotonous task – as mentioned in the previous example – and thus contribute to a socio-economic change in our working environment (Precht, 2022), but can also reduce greenhouse gas emissions through more resource-efficient systems: From intelligent power grids and transportation systems to automated production in “smart cities” (Clutton-Brock et al., 2021).



References:


Clutton-Brock P., Rolnick D., Donti P.L., Kaack L.H.: Climate Change and AI. Recommendations for Government Action, 2021. https://gpai.ai/projects/responsible-ai/environment/climate-change-and-ai.pdf


Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun: Deep Residual Learning for Image Recognition, 2015.


Ronneberger O., Fischer P., Brox, T.: U-Net: Convolutional Networks for Biomedical Image Segmentation, 2015.


Precht R.D.: Freiheit für alle: Das Ende der Arbeit wie wir sie kannten, 2022.


Yunwei Tang, Fang Qiu, Bangjin Wang, Di Wu, Linhai Jing & Zhongchang Sun: A deep relearning method based on the recurrent neural network for land cover classification, 2022. GIScience & Remote Sensing, 59:1, 1344-1366, DOI: 10.1080/15481603.2022.2115589

error: Dieser Inhalt ist geschützt!
Cookie Consent Banner by Real Cookie Banner