**5. Conclusions**

This study proposed a paddy rice mapping approach based on a weakly supervised LSTM network with DTW distance-based sampling strategy. The purpose of this study was to enable a good fit for deep learning classifiers in the circumstance that field samples were not sufficient or to reduce the cost of field sampling. Apart from field samples, a larger number of weak samples were labeled on the basis of DTW distance to the standard SAR profiles of each land cover type. A specifically designed LSTM classifier was trained on combined optical spectral bands and SAR backscatter data as input features. A paddy rice map was finally generated with the best training scheme.

A few conclusions can be made on the basis of experiment results. First, the weakly supervised approach has a positive impact on paddy rice mapping when limited field samples are available. The DTW distance-based sampling strategy effectively reduced field sampling costs since fewer field samples were required. Second, the LSTM network classifier achieved high precision on paddy rice mapping. The mechanism of LSTM networks is feasible for paddy rice mapping where variance exists in terms of planting and harvesting schedules.

**Author Contributions:** Conceptualization, M.W.; data curation, L.C.; methodology, M.W.; project administration, J.W.; resources, L.C.; software, M.W.; writing—original draft, M.W.; writing—review and editing, J.W. and L.C. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was supported by the China Central Public-Interest Scientific Institution Basal Research Fund (no. Y2020PT14), Basal Research Fund of AII CAAS (no. JBYW-AII-2020-17), Open Research Fund Program of Key Laboratory of Agricultural Big Data, Ministry of Agriculture and Rural Affairs, and Special Project of National Science and Technology Library (no. 2020QBW008).

**Conflicts of Interest:** The authors declare no conflict of interest.
