Phenology-Based Unsupervised Rapeseed Mapping Using Multitemporal Data

With a large field of view, remote sensing is a useful technique that can provide a new way to acquire crop acreage and spatial distribution. As an important oil crop, rapeseed is widely planted in many countries. Besides, rapeseed mapping is of great significance for food security and policy regulation. The objective of this work is to explore an automatic and effective framework for rapeseed mapping using multitemporal multispectral images (MSIs) captured by multiple sensors. The automatic rapeseed mapping framework consists of two stages. Concretely, color transfer is introduced and combined with vegetation indexes to detect rapeseed for the first time. After that, the most general supervised classification methods are applied to optimize the initial mapping result. In the experiments, the performance of different methods is evaluated on multisensor data within four years (2017–2020). Moreover, to analyze the effect of temporal information, the mapping performance of using different MSIs as input is systematically compared. Experimental results illustrate the effectiveness of the proposed framework, in which color transfer makes an important role. Some valuable findings are also obtained which can be further used for global rapeseed mapping.