Selection of unlabeled source domains for domain adaptation in remote sensing

—In the context of supervised learning techniques, it can be desirable to utilize existing prior knowledge from a source domain estimate target variable in by exploiting concept adaptation. This is done alleviate costly compilation knowledge, i.e., training data. Here, our goal select single for adaptation multiple potentially helpful but unlabeled domains. The data solely obtained if was identified as being relevant estimating corresponding selection mechanism. From methodological point view, we propose unsupervised voting (an ensemble of) similarity metrics that follow aligned marginal distributions regarding image features and Thereby, also an pruning heuristic include robust scheme. We provide evaluation methods models sets created with Level-of-Detail-1 building regress built-up density height on Sentinel-2 satellite imagery. To evaluate capability, learn apply interchangeably four largest cities Germany. Experimental results underline capability obtain more frequently higher accuracy levels improvement up 10% most mechanisms compared random source-target selections.