Exploration of using satellite SAR data for crop classification

• Main Author: Sim, Yee Fei
• Other Authors: Lu Yilong
• Format: Final Year Project
• Language: English
• Published: 2019
• Subjects: DRNTU::Engineering::Computer science and engineering::Computing methodologies::Image processing and computer vision
• Online Access: http://hdl.handle.net/10356/77847
• Institution: Nanyang Technological University

Crop classification is one of the most imperative tasks in the agriculture field to meet rising food demand for most of the world’s growing population. Therefore, accurate and up-to-date evaluation of the temporal and spatial distribution of crop cultivated area are key issues for agricultural monitoring. Remote sensing techniques is used to evaluate the effect of agricultural land cover on the ecosystems. However, Sentinel-2 data provide accurate crop classifications only if the cloud free acquisitions time series is large enough and if the acquisition dates are taken in the good agricultural period. But when the study area is very cloudy, the number of available optical images could be insufficient for crop classification. In 2014, European Space Agency (ESA) launched the satellite Sentinel-1 with synthetic aperture radar (SAR) for the Copernicus program. SAR data is particularly appealing to crop classification due to its high-resolution capability, which is independent of weather and great all-day imaging capability. Therefore, the purpose of the project is to explore the capability of using optical fusion and SAR data to classify crop types. SAR time series from Sentinel-1 is used to combine with the optical Sentinel-2 to improve the performance. The project is carried out using 15580 training data situated in northern France. The results indicated that the Sentinel-2 images compounded with Sentinel-1 time-series data enabled high classification accuracies of winter sweet wheat, non-fodder beet and rape (F1-score above 0.8) using Random Forest classification method.