M.Sc. thesis handed in by Marcus Groll

M.Sc. thesis handed in by Marcus Groll

Marcus Groll handed in his M.Sc. thesis “Deep learning for Instance Segmentation of bomb craters on historical aerial images of the Second World War ” and will defend it next week.

Abstract: During the Second World War (WWII) many air strikes were flown on German cities. A long time after the war, the consequences can still be recognized. About ten to fifteen percent of the dropped bombs are still lying unexploded in the ground. To find these unexploded ordnances (UXO) is nowadays an important task. Most of the UXO searching is done manually. An automatic segmentation of bomb craters on historical aerial images would help to simplify the search. So, the goal of this thesis is the creation of a deep learning instance segmentation of bomb craters from historical aerial images of WWII. The masked region based convolutional neural network (Mask R-CNN) application by ABDULLA (2017) was trained on the historical images from all over Germany. It is also possible to use this concept on every historical aerial image worldwide. The Mask R-CNN method was published in the early 2017 and until now never used for the detection of bomb craters on historical aerial images. DUTTA ET AL (2016) did the collection of the training polygons with the VGG Image Annotator. In summary 50 training and 30 validation data were used. It was decided to collect five validation images for each setting. The final Mask R-CNN model is validated via the comparison of the influences by different image settings on the model. As the image settings were chosen weather situations, bombing methods and the recorded image sizes. The TensorBoard approach and the calculation of the Average precision (AP) was applied for the validation of the model. In addition to the influence by every setting, the effects by the images themselves, hardware and software are also summarized. Finally, every setting got a big influence on the result of the model, especially the hardware and software setting. The right hardware- and software setup is a very important preprocessing step to prevent “MemoryError” messages by the system. The company Luftbilddatenbank (LBDB), in Estenfeld, close to Würzburg, Germany, provided all the historical images. This thesis was written in cooperation with them. A further target of this thesis is the creation of a plugin in a Geographic Information System (GIS) software like ArcMap or QGIS, which is able to do the instance segmentation of bomb craters. This plugin should be implemented into the workflow of the LBDB.

Maninder Singh Dhillon handed in his MSc

Maninder Singh Dhillon handed in his MSc

Maninder Singh Dhillon handed in his M.Sc. “Comparing the performance of crop growth models using synthetic remote sensing data at DEMMIN, Germany” supervised by Thorsten Dahms with the first and second supervisors Martin Wegmann and Christopher Conrad, respectively.

His M.Sc. defense will be on Monday 18th at 2pm in room 0.004 (Oswald-Kuelpe Weg). Read the abstract of his M.Sc. for more details about his research project.

 

 

 

Abstract: Climate change, natural resource degradation, and population growth are increasingly challenging food security. Climatic effects such as warming temperatures, and decrements in soil moisture have already stagnated the winter wheat yields in many parts of the world. The present study was conducted over three years from 2015 to 2017 on a main agricultural production region of North-East Germany. Five widely used crop models, namely, WOFOST. CERES, Aqua crop. LUE and CropSyst, are compared to predict the biomass of winter wheat crop. The study stresses the use of remote sensing with crop growth models, as they lack spatial information on the actual conditions of each field or region. It also elaborates the concept of data fusion, in which both Landsat (30 m) and Modis (500 m) are combined to get synthetic remote sensing data of 30-meter spatial and one-day temporal resolution, in crop growth modeling with its contribution to crop growth models. Measurements of the biophysical parameters such as Leaf Area Index (LAI), Canopy Cover (CC), Fraction of Photosynthetically Absorbed Radiation (FPAR) were grouped according to the phenology using the BBCH characterization (Biologische Bundesanstalt, Bundessortenamt und Chemische Industrie). The analysis was carried out for winter wheat during the years 2015 to 2017 and included a detailed comparison of the simulated and measured crop biomass as well as an impact of climate parameters such as temperature stress, soil moisture stress and vapor pressure deficit (VPD) on crop growth, and an evaluation of the modelled crop biomass. The study shows that the models (Aqua crop, CropSyst, and LUE) that require fewer input parameters to simulate crop biomass are well suitable and easier to implement.

Internship and M.Sc. idea presentations

Internship and M.Sc. idea presentations

on Thursday, December 13th, at 12:30 we will have the following presentations in the student working room (Josef Martin Weg 52, 3rd floor):

internship presentations:

 
Johni Miah
“Remote Sensing and Geographic Information System for Decision Making”
 
Benjamin Lee
“Quantifying intertidal areas in the East Asian-Australasian Flyway”
 
Marina Reiter
“Computation of sewer system in Ulm”

M.Sc. idea presentations:

 
Marina Reiter
“Analysis of urban green areas in German cities”
 
Fowad Ahmed
“Drought Monitoring in the University Forest Sailershausen”

M.Sc. graduation by Jakob Schwalb-Willmann

M.Sc. graduation by Jakob Schwalb-Willmann

Congratulation to Jakob Schwalb-Willmann who successfully graduated today! His M.Sc. topic was “A deep learning movement prediction framework for identifying anomalies in animal-environment interactions” aiming to explore the potential of animal movement analysis for informing remote sensing data analysis. 

The M.Sc. was conducted in collaboration with the Max-Planck-Institute of Ornithology and supervised by Martin Wegmann and Kamran Safi.

From the abstract: “The environmental conditions that animals are exposed to infuence their movement throughout the landscape. A variety of modeling approaches aim to quantify the relationships between animal movement trajectories and environmental variables, e.g. acquired through satellite remote sensing. However, many of such approaches are designed for species-specifc movement modeling and often rely on a priori assumptions, knowledge-based parameterization or the selection of features. In this study, the use of representation learning for predicting animal-environment interactions in a non-parametric, species-independent framework is investigated to enable the unsupervised detection of anomalies. It is shown that a Long Short-Term Memory (LSTM) network is capable of learning generalized representations of the interactions between location and raw environmental features in movement sequences which had been simulated under an agent-based resource selection regime unknown to the network. Thus, the network is able to reconstruct such movement sequences that show similar feature interaction characteristics to the ones the network had learned. On the contrary, the reconstruction of sequences containing patterns that are novel to the network results in a reconstruction error response that was utilized to detect anomalies in movement sequences. It is discussed, how the proposed framework may be applicable for the analysis of animal movement, e.g. for outlier detection, behavioral segmentation or movement simulation. In addition, its potential for supporting environmental research on an inter-species scale, e.g. by detecting phenological indicators from animal movement, is examined.

EAGLE M.Sc. idea presentations

EAGLE M.Sc. idea presentations

On Monday, 24th of September from 1:30 onwards the following EAGLE students will present their M.Sc. idea. Everybody is welcome to join their presentations and to provide feedback:

Julia:
“Time-Series Analysis of Sentinel-1 and Sentinel-2 Imagery for Detection of active Morphodynamics in the Atacama Desert, Chile”

Sarah:
Risk assessment for flood events based on remote sensing data – a case study for North-Rhine Westfalia, Germany

Louis:
“Remote sensing of water quality using Sentinel-2 towards a potential separation of harmful algal blooms from other Algae”

Bharath:
“Assessing the development of circular irrigation in South Africa since 19*”

Johannes:
“Tracking crop penology based on S1-Time series”

Sebastian:
“Solar power potential in Portugal”

Karsten:
“An approach to optimize object-based classification: Mapping dead trees and gaps in Białowieża Forest”

Marcus:
“Deep learning for Instance Segmentation and a comparison to a Conventional Segmentation on historical aerial images of the Second World War”

Fowad:
“Drought Monitoring in University Forest with Hyper-Spectral And In-Situ Data”

New MSc thesis: Time series analysis in Colombian Orinoco Basin

New MSc thesis: Time series analysis in Colombian Orinoco Basin

Pilar Endara started her M.Sc. thesis on “Time series analysis of flooding and vegetation patterns in wetlands of the Colombian Orinoco Basin”

The ecosystems that are present within Colombian Orinoquia flooded savannas are currently being threatened by conversion of natural systems into intensive rangelands with introduced pastures, croplands and palm oil plantations. The loss of natural floodplain ecosystems has serious negative impacts on several important ecosystem services such as habitat quality for biodiversity, long-term carbon sequestration and water regulation.

In addition, this region’s natural vegetation endures strong intra-annual hydrological regimes, which in turn may affect the productivity, and consequently, the region’s carbon dynamics. Therefore, the objective of this thesis is to understand how, and if, the vegetation productivity in this area varies along the year, taking into account the strong hydrological variability.

The results of this thesis are going to be integrated with information from WWF Colombia to provide stakeholders and decisions-makers a tool that will allow them to assess for best practices of land use planning and land management.

First supervisor is Claudia Künzer, second supervisor is Martin Wegmann