Jakub Nalepa

Identifying methane plumes in Earth observation image data is vital in environmental monitoring. This process is the key to locating methane emissions, which is critical for efforts aimed at reducing their contribution to climate change. However, the potential overlaps between methane and other gases in their spectral features make the identification of methane emitters challenging. Therefore, new, more effective and efficient algorithms are needed to address this issue. We propose incorporating methane enhancement maps extracted from image data using improved matched filtering into the feature vectors that undergo evolutionary selection before feeding them to supervised classification models. Therefore, to tackle the problem of methane detection and localization in hyperspectral images, we use a framework that combines evolutionary techniques for feature selection, followed by supervised machine learning for classifying hyperspectral superpixels as those containing or not containing methane. The experiments over the STARCOP benchmark dataset show that even if we exploit no more than three features, we obtain models whose quality, quantified as the area under the receiver operating curve, reaches 0.885 (with a sensitivity of 0.880). For reproducibility, we refer to our GitHub repository: https://github.com/smile-research/GeneticAlgorithms4MethaneDetection.git.

Technological advancements are expanding the potential of hyperspectral image (HSI) analysis for Earth observation. Extracting insights from high-dimensional images led to various supervised artificial intelligence (AI) approaches. However, the world is not labeled, and acquiring ground truth (GT) is expensive. For certain tasks, such as estimating soil parameters, only coarse-grained, field-level measurements are available. We tackle the challenge of building (un)supervised AI models from weakly-labeled sets with image-level labels. We propose a comprehensive framework for estimating soil parameters from HSIs, and a spectrally- and spatially-informed algorithm for generating pseudolabels based on the original GT. We analyze the spatial variations of spectral pixel characteristics within the parcels (images) using superpixels, which group neighboring pixels based on their spectral features—superpixels are determined using unsupervised clustering. Then, the superpixels are clustered (based on the feature vectors calculated for the superpixels). The resulting clusters are either used to estimate soil parameters for the incoming unseen samples (pixels, superpixels, or images) in a fully unsupervised fashion, or are exploited to elaborate pseudolabels (at the superpixel level) which can be used to train supervised models. The experiments not only showed that our (un)supervised methods outperform the supervised state-of-the-art models on the HYPERVIEW benchmark, collocating HSIs and in-situ measurements of soil parameters, but also indicated that pseudolabels make the modeling process much easier and generalizable, and may be considered as de-noising of the originally acquired data. We release our models and pseudolabels to ensure reproducibility of our study.

The analysis of satellite images has attracted significant research interest due to its numerous applications and unparalleled scalability in Earth observation (EO). Although artificial intelligence (AI) algorithms for EO emerge at a steady pace, the community still needs to address many practical challenges that are concerned with such highly dimensional and unprecedentedly large volumes of image data. Quantum machine learning (QML) is a promising research avenue here. Despite the growing interest and funding in the field, current results remain inconclusive, shifting focus toward understanding the strengths and limitations of QML rather than solely expanding the applications of such methods. This article addresses a critical gap by consolidating advancements in QML, with a special focus on quantum kernel methods—which already proved their value in EO—to evaluate their role in advancing state-of-the-art (SOTA) EO solutions and exploring the potential quantum advantage via identifying their benefits and shortcomings in an unbiased and thorough way.