1 May 2023 in Journal of Spacecraft and Rockets
9 May 2024 in IEEE Geoscience and Remote Sensing Magazine
Enhancing agricultural methods through the utilization of Earth observation and artificial intelligence (AI) has emerged as a significant concern. The ability to quantify soil parameters on a large scale can play a pivotal role in optimizing the fertilization process. While techniques for noninvasive estimation of soil parameters from hyperspectral images (HSIs) exist, their validation typically occurs across different datasets and employs varying validation protocols. This diversity renders them inherently challenging (or even impossible) to compare objectively.
1 September 2023 in Materials Research Proceedings
Pose estimation from a monocular camera is a critical technology for in-orbit servicing missions. However, collecting large image datasets in space for training neural networks is impractical, resulting in the use of synthetic images. Unfortunately, these often fail to accurately replicate real image features, leading to a significant domain gap. This work explores the use of generative adversarial networks as a solution for bridging this gap from the data level by making synthetic images more closely resemble real ones. A generative model is trained on a small subset of unpaired synthetic and real pictures from the SPEED+ dataset. The entire synthetic dataset is then augmented using the generator, and employed to train a regression model, based on the MetaFormer architecture, which locates a set of landmarks. By comparing the model's pose estimation accuracy on real images with and without generator preprocessing, it is observed that the augmentation effectively reduces the median pose estimation error by a factor 1.4 to 5. This compelling result validates the efficacy of these tools and justifies further research in their utilization.