The M.Sc. thesis was resarched, written and defended at the University of South Brittany in Vannes, France between Februrary and June 2022.
Feature Extraction via Transfer Learning and using predefined Model Architectures for Palm Plantation Identification
Several predefined deep learning computer vision models, such as LeNet, ResNet18, VGG16 and AlexNet are used to identify palm oil plantations in satellite images. The models are both trained from scratch and used as a feature extractor via transfer learning on the pretrained models.
Performing some Fourier Transformations and filtering operations on images.
Performing edge extraction, Harris corner point identification and Hughes line detection. Additionally, a Haralick texture analysis is performed and a k-Nearest Neighbor approach tested basted on the texture information.
Performing different matrix manipulations to geometrically alter images and reproject the color in different color spaces.
Following an UAV data acquisition, orthoimages for RBG, RE, and NIR bands as wel as a DSM are calculated. Different combinations of these images are fed into a classification algorithm to test how important each band is for the resulting classification.
Report of Internship at German Aerospace Center, completed in 2021.
Based on the global marine traffic dataset, several MapReduce functions are implemented in Python via ‘mrjob’.
. In this notebook, Gaussian Process Regression is performed based on perfect and noisy data in order to show the basic principles behind the process.
In this project, two point clouds of the same suburb near Christchurch in New Zealand are loaded into CloudCompare and the difference in altitude is calculated.