I would be happy to support you in designing training and test datasets for machine learning. In addition, I also offer you the possibility to supplement your datasets with newly generated and open-source data – for example, photo data, graphics datasets using CGI, artificial training datasets using GAN techniques, and satellite imagery.
I also run a photo archive for geo analytical questions, illustrating the climatic landscape changes in our geosystem earth. Don’t hesitate to contact me if you want to use photographs for social or scientific purposes.
It is usually helpful to extend the already available training data by data augmentation to train a photo classification method with computer vision techniques. This can significantly improve the accuracy of the image classification process. To generate new photo data, I either use existing techniques from the Python libraries Tensorflow and Pytorch or develop a specially designed artificial neural network for you.
CGI techniques are suitable for augmenting training data for machine learning methods and thus influencing the accuracy of image classification. In addition to manual data generation with Blender, Daz3D and GrassGIS, for example, existing datasets can also be used. I will be happy to advise you in this regard and provide CGI data if necessary.
With the help of Generative Adversarial Networks (GAN), i.e. a deep learning algorithm, artificial training data for machine learning processes can be generated. For unbalanced datasets, this data augmentation can be beneficial and improve the accuracy of existing machine learning techniques. I am happy to develop and train an artificial neural network (GAN) for you to generate additional training data.
It usually makes sense to supplement the existing training data with data augmentation to train image segmentations based on satellite images using computer vision techniques. This can usually significantly improve the segmentation accuracy of the machine learning process. Furthermore, complementary images from different satellites from different time series can improve accuracy. I would be happy to support you in acquiring satellite images, including training a specially developed artificial neural network.
Photographic comparisons of landscapes over different periods give us clues about the anthropogenic and physiogeographical changes on the earth's surface.
The photographs are indispensable for future generations - especially in the context of our anthropogenically induced climate change. This is precisely why continuous documentation of the earth's surface is relevant.
Take a look at the photo archive and immerse yourself in a world of remote and impressive mountains and polar regions.