CORINE Land-Cover (CLC) and its by-products are considered as a reference baseline for land-cover mapping over Europe and subsequent applications. CLC is currently tediously produced each six years from both the visual interpretation and the automatic analysis of a large amount of remote sensing images. Observing that various European countries regularly produce in parallel their own land-cover country-scaled maps with their own specifications, we propose to directly infer CORINE Land-Cover from an existing map, therefore steadily decreasing the updating time-frame. No additional remote sensing image is required. In this paper, we focus more specifically on translating a country-scale remote sensed map, OSO (France), into CORINE Land Cover, in a supervised way. OSO and CLC not only differ in nomenclature but also in spatial resolution. We jointly harmonize both dimensions using a contextual and asymmetrical Convolution Neural Network with positional encoding. We show for various use cases that our method achieves a superior performance than the traditional semantic-based translation approach, achieving an 81% accuracy over all of France, close to the targeted 85% accuracy of CLC.

This paper presents a framework for simultaneously translating multiple land-cover maps into a given one in a supervised way. Conversely to existing approaches working on 1-1 translation, we propose a multi-translation setup that increases the generalizability and translation performance, especially on land-cover maps covering restricted spatial extents. The proposed method mainly assumes that the map of interest spatially overlaps at least with one of the other maps. High performance translation is achieved with a Convolutional Neural Network (CNN) based encoder-decoder framework trained with three goals: (i) high-quality translation; (ii) self-reconstruction ability; (iii) mapping of all datasets into a common representation space. Country-scale experimental results show the method effectiveness in translating six highly heterogeneous land-cover maps, achieving significantly better results than the traditional semantic-based method and better results than CNN trained for a 1-1 translation task (+ 9.7% in Overall Accuracy (OA) and +12% in macro F1-score (mF1)).

No abstract

No abstract