Quantitative Wood Anatomy (QWA) anomaly detection
Quantitative Wood Anatomy (QWA) analyzes multiple anatomical features of microscopic wood images of thin sections. These features can be directly related to environmental influences during tree growth. For this analysis, specialized software detects the lumen area of tree cells. Recently different machine learning tools emerged which increased the performance [1,2]. A still challenging and untackled problem is to distinguish between anomalies in the wood and the analyzable structures. Therefore we aim to train a deep neural network for segmenting different anomalies.
Detailed description: Quantitative Wood Anatomy (QWA) anomaly detection (PDF, 2 MB)
Tree ring detection for Quantitative Wood Anatomy (QWA)
Quantitative Wood Anatomy (QWA) analyzes multiple anatomical features of microscopic wood images of thin sections. These features (e.g., tree rings) can be directly related to environmental influences during tree growth. In current software, tree rings and other features are inferred from the detected cells, which is error-prone if the detection algorithm misses the smallest cells that are hardest to detect. We thus aim to use a deep neural network to directly identify tree ring borders without reliance on cell segmentation. Additionally we want to investigate if other features e.g. cell wall thickness can be directly estimated.
Detailed description: Tree ring detection for Quantitative Wood Anatomy (QWA) (PDF, 1 MB)
Refinement of Large-scale Vegetation Height Map with Self-Supervised Learning
Although traditional field measurements combined with airborne laser scanner (ALS) can produce accurate vegetation height map (VHM) with fine details, it is infeasible to generate a global map with ALS due to its high time and labor cost. The recent satellite mission with spaceborne LiDAR instruments (GEDI [Dubayah et al., 2020]) provides sparse vegetation height measurement across a large part of the earth and enables the generation of global VHM. Recent works [Lang et al., 2022a,b] have successfully used deep learning techniques to generate global VHM with 10-meter resolution based on Sentinel-2 images and GEDI data. However, the discrepancy of resolution between the input image and the reference data poses a problem (check the relationship of resolution between input and reference data in Figure 1a): models trained on GEDI reference data with 25-meter footprint yield a reduced effective resolution. Therefore, we aim to use a self-supervised method to recover the small texture details omitted in the current global VHM map. As shown in Figure 1b, we plan to formulate this problem as a guided superresolution problem: use the 10-meter Sentinel-2 image as a guide and the coarse VHM as a source image to produce a refined VHM that retrains more highfrequency
details. The refined map should contain the vegetation height information from the source image and inherit the fine texture details from the guide image.
Bayesian Deep Learning for crop classification – measuring uncertainty in Remote Sensing predictions
To map tropical crops like oil palms or cocoa, uncertainty in the model's prediction must be quantified and considered for actual sustainable growth, production and supply chains. Yet, map uncertainty is far from being standard in applied deep learning systems. In this thesis, we want to explore the applicability of current methods to crop classification and segmentation in order to produce uncertainty maps alongside standard classification maps. The goal of this thesis is to implement existing Bayesian deep learning frameworks to investigate their applicability to crop classification in Remote Sensing. As a first step, the student would need to get a coarse overview of current state-of-the-art methods in Bayesian deep learning and evaluate their applicability to crop classification. Then, the most promising approach can be implemented in our existing deep learning pipeline with input from satellite imagery and different crops as ground truth. We would start with a binary segmentation task that can be easily extended to multiclass tasks. Finally, the student would evaluate the uncertainty measures both qualitatively and quantitatively. Prerequisites are a strong interest in deep learning as well as good programming skills in Python and common machine learning libraries e.g. PyTorch.
Contact: Prof. Prof. Jan Dirk Wegner
Functional diversity mapping in agroforestry - linking in-situ data and satellite observations
Biodiversity is globally at threat by many global change related drivers and the in-situ monitoring of changes in biodiversity through observational networks is costly and very difficult to implement for larger regions. Remote sensing (RS) data could fill the spatiotemporal gaps in such observational networks and provide wall-to-wall information on trends in biodiversity change, potentially informing decisions on mitigation strategies. However, such approaches need to be validated and tested across biomes before they can serve such tasks. In this thesis, a recently developed approach for the mapping of forest functional diversity based on Sentinel 2 multi-spectral data shall be transferred to central Africa (Ghana and Cote-d-Ivoire), where it can be validated with in-situ data (biomass and species diversity) and close-range remote sensing data (UAV based orthoimages and surface models).
Detailed description: Functional diversity mapping in agroforestry (PDF, 320 KB)
Diversity in the low mass galaxy population
Observations show that low-mass galaxies fall into several broad morphological classes such as ultra-diffuse galaxies, dwarf elliptical galaxies, compact elliptical galaxies, and others. Current numerical models struggle in reproducing this diversity in the low mass galaxy population. The goal of this project is to more robustly quantify this problem and to explore improvements to the modeling to solve this challenge. To this end, we will analyze low-mass galaxies in suites of state-of-the-art cosmological simulations and compare the predicted morphologies, especially their sizes and rotation curves, faithfully with observational data of nearby low-mass galaxies.
Contact: Prof. Robert Feldmann
Disk galaxy formation
When and how galaxies form stable, thin disks is still not fully understood. A recent model proposes that disk formation is closely linked to the depth of the gravitational potential in the center of galaxies. First, we will test this model prediction and explore its limitations using a large suite of cosmological simulations. Secondly, we will analyze the properties of both the gas and stellar disks in more detail and compare them to available observational data. Finally, we will make predictions for when the first disk galaxies are expected to form in the Universe.
Contact: Prof. Robert Feldmann
Inferring galaxy properties with Machine Learning
Galaxy properties, such as their masses or star formation rates, are usually observationally inferred from the magnitudes and colors of galaxies. However, interstellar dust can strongly affect these measurements. In many cases, the impact of dust is modeled in a highly simplified and idealized manner which could in principle lead to significant biases in the inferred galaxy properties. In this project, we will use galaxies from a cosmological simulation which have been processed with radiative transfer to quantify these biases and to test how well we can recover intrinsic galaxy properties. As second part, we will train a deep neural network on the simulations to predict intrinsic quantities from mock images and apply it to a real observational data set.
Contact: Prof. Robert Feldmann
Holes in atomic gas disks
Gas disks of galaxies often show distinct holes which may be created by stellar feedback, turbulence in the interstellar medium, and other processes. The goal of this project is to study holes in the atomic gas disks of simulated galaxies, to quantify their properties, and to determine how they formed. In addition, we will develop a machine learning based classifier to automatically distinguish the different physical origins of holes in atomic gas disks.
My group works with N-body simulations of cosmic structure formation, ray tracing and statistical analysis of simulation data. We can offer projects with the aim to extract statistical features from the distribution of matter in the Universe. Examples include
- A project to develop improvements to the two-point correlation function estimator for large, high-quality data sets
- A project to study the effect of relativistic frame-dragging on the statistics of gravitational weak-lensing tangential shear
Contact: Prof. Julian Adamek
Giant planet formation and evolution
Giant planets play a key role in the formation and dynamical evolution of planetary systems. Their origin, however, is still poorly understood. In this project we will use numerical simulations to model the formation and evolution of giant planets with the aim to provide predictions on their compositions and final masses of the planets. We will explore the formation of the outer planets in the solar system as well as of giant exoplanets.
Contact: Ravit Helled
Planets orbiting other stars are detected at a rapid pace. When both the mass and the radius of a planet are measured, the average density of the planet can be determined. In order to understand what exoplanets are made of, structure models are required. In this project we will characterise exoplanets using interior models, to put limits on their bulk compositions and internal structure, and when possible, link it to their formation and evolution histories.
Contact: Ravit Helled
Giant Impacts of clumps
Protoplanetary disks that are massive enough can fragment leading to the formation of clumps that could evolve to become giant planets. Simulations suggest that several clumps form in such disks. The interactions between these clumps and their potential collisions that can lead to both mergers and fragmentation are poorly known. In this project we plan to simulate the collisions between gaseous clumps using SPH simulations and determine their fate for a large range of initial and impact conditions.
Contact: Ravit Helled & Joachim Stadel
The dipole of the Universe
The CMB observations provide a precise measurement of the dipole, which is due to the motion of our local system with respect to the CMB rest-frame. Other cosmological probes can be used to measure the dipole, and the consistency check of our cosmological models can be made in comparison to the CMB observation. Supernova observations and galaxy surveys are widely used to measure the dipole. A more rigorous analytical investigation of these methods would reveal the nature behind the observational discrepancy.
Contact: Prof. Jaiyul Yoo
Covariant model of galaxy bias
In literature galaxy bias models are developed in the Newtonian framework. However, the symmetry of GR is the diffeomorphism symmetry, and the current galaxy bias model is incompatible with the diffeomorphism symmetry. The Newtonian description works well in most cases, but not always. One of the key target in the upcoming large-scale surveys is to measure the primordial non-Gaussian signal on very large scales, at which the relativistic effects are important. A correct covariant model of galaxy bias is needed to take advantage of the full potential in the upcoming large-scale surveys. One project is to generalize the peak model in the Newtonian dynamics to a covariant model in the relativistic dynamics. The other project is to explore the impact of a simple covariant toy model in terms of renormalization.
Contact: Prof. Jaiyul Yoo