Bachelor and Master Theses Propositions in Remote Sensing

Focus Areas

  • Land use and land cover classification
  • Pattern recognition in large data sets
  • Classification methods, such as supervised learning, unsupervised learning, self-taught learning
  • Feature learning / representation learning
  • Analysis techniques, such as clustering or archetypal analysis
Determination of Sea Surface Heights using Retracking and Shortest-Path Algorithm

German title: Bestimmung der Meereshöhen durch Retracking und Kürzeste-Wege Algorithmus

Satellite radar altimetry is one of the most powerful techniques for measuring sea surface height variations, with applications ranging from operational oceanography to climate research. Over open oceans, altimeter return waveforms generally correspond to the Brown model, and by inversion, estimated shape parameters provide mean surface height and wind speed. However, the estimated sea surface heights depend highly on the used retracking algorithm and its chosen hyperparameters, resulting in multiple likely heights for each position.

The goal of the thesis is the development of an efficient shortest-path algorithm to find a best set of sea surface heights from multiple likely heights. For the experiments, estimated sea surface heights obtained from the STAR retracker [1] will be used. The study sites cover the Gulf of Trieste, Italy, and the coastal region of the Ganges-Brahmaputra-Meghna estuary, Bangladesh

[1] Roscher, R., Uebbing, B., & Kusche, J. (2017). STAR: Spatio-temporal altimeter waveform retracking using sparse representation and conditional random fields. Remote Sensing of Environment, 201, 148-164.

Supervision: Ribana Roscher, Jan-Henrik Haunert

Grapevine Berry Detection using Convolutional Neural Networks

German title: Detektion von Beeren mittels Convolutional Neural Networks

The berry size is one of the most important fruit traits in grapevine breeding. Non-invasive, image-based phenotyping promises a fast and precise method for the monitoring of the grapevine berry size.

The goal of this thesis is the development of an automated image analyzing framework for the detection of grapevine berries from images. For this, convolutional neural networks will be utilized. They state a powerful approach from the deep learning area, which are able to capture spectral as well as spatial features from the images. The experiments will conducted using RGB images acquired by a robot in a vineyard.

Supervision: Ribana Roscher

Probabilistic classifiers for detection of unknown classes in multispectral remote sensing data

German title: Probabilistische Klassifikation für die Detektion unbekannter Klassen in multispektralen Fernerkundungsdaten

Classification is one of the most informative means for interpreting remote sensing images. One challenge is the interpretation of areas with unknown land use and land cover classes, i.e. they are not included in the classification model. In order to ensure a reliable and certain classification result, these areas need to be identified by means of suitable techniques. Probabilistic classifiers with reconstructive model component are able to detect these areas. The goal of the thesis is the assessment of the suitability of different probabilistic classifiers for remote sensing data classification. The evaluation will be conducted on various multispectral datasets with differing characteristics.

Supervision: Ribana Roscher, Susanne Wenzel

Sparse Representation for root segmentation in hyperspectral images

root image in rhizotrone

German title: Sparse Representation für die Segmentierung von Wurzeln in Hyperspektralbildern

The segmentation of roots in images is challenging due to their similar color appearance and their thin structure. Sparse representation can be used for segmentation by representing each image patch by a weighted linear combination of basis patches (e.g., [2]). These basis patches are learned from annotated images and will be collected in discriminative dictionaries, which contain the most distinct structures of roots and background. The suitability of the approach will be evaluated by means of hyperspectral images of plants grown in flat rhizotrons [1].

[1] Steier, A., Cendrero-Mateo, P., Malolepszy, T., Gioia, T., Briese, C., & Rascher, U. (2013). HIDeR – Hyperspectral Imaging Device for Rhizotrons Simultaneously Scanning of Shoot and Root in Plant Phenotyping, DPPN.
[2] Berkels, B., Kotowski, M., Rumpf, M., & Schaller, C. (2011, May). Sulci detection in photos of the human cortex based on learned discriminative dictionaries. In International Conference on Scale Space and Variational Methods in Computer Vision, 326-337.

Supervision: Ribana Roscher, Uwe Rascher

Incremental learning on large image mosaics

German title: Inkrementelles Lernen auf großen Bildverbänden

Large image mosaics from remote sensing images are characterized by spatial and temporal differences, and thus spectral differences. Therefore, the interpretation of large image mosaics is challenging. Especially if the number of labeled samples with information of land use and land cover is limited, the learning of a well generalized classification model is difficult. In contrast to batch-based learning methods, incremental learners are able to train a classification model which adapts to the given data over time. Goal of this thesis is the classification of an image mosaic using incremental learning. A main part of the thesis will consider the acquisition of new data samples in order to adapt the classification model to current conditions.

Supervision: Ribana Roscher, Susanne Wenzel