2019/11/20 by Yawogan Jean Eudes Gbodjo, Dino Ienco, Gbodjo, Yawogan Jean Eudes +11
Computer Science · Engineering · Environmental Science · #FOS: Computer and information sciences #Geochemistry and Geologic Mapping #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Remote Sensing in Agriculture #Remote-Sensing Image Classification
paper · pdf · doi:10.48550/arxiv.1911.08815
openalex publication_date 2019/11/20 · openalex created_date 2022/07/22 · openalex updated_date 2026/07/28
European satellite missions Sentinel-1 (S1) and Sentinel-2 (S2) provide at\nhighspatial resolution and high revisit time, respectively, radar and optical\nimagesthat support a wide range of Earth surface monitoring tasks such as\nLandUse/Land Cover mapping. A long-standing challenge in the remote\nsensingcommunity is about how to efficiently exploit multiple sources of\ninformation and leverage their complementary. In this particular case, get the\nmost out ofradar and optical satellite image time series (SITS). Here, we\npropose to dealwith land cover mapping through a deep learning framework\nespecially tailoredto leverage the multi-source complementarity provided by\nradar and opticalSITS. The proposed architecture is based on an extension of\nRecurrent NeuralNetwork (RNN) enriched via a customized attention mechanism\ncapable to fitthe specificity of SITS data. In addition, we propose a new\npretraining strategythat exploits domain expert knowledge to guide the model\nparameter initial-ization. Thorough experimental evaluations involving several\nmachine learningcompetitors, on two contrasted study sites, have demonstrated\nthe suitabilityof our new attention mechanism combined with the extend RNN\nmodel as wellas the benefit/limit to inject domain expert knowledge in the\nneural networktraining process.\n