Incorporating Phenological Patterns and Multi-source Remote Sensing Imagesfor Cropland Change Detection

Abstract

Accurate information on cropland changes is critical for monitoring arable land minimum, ensuring national food security, and grasping the situation of agricultural production and supply. The change detection using remote sensing images is one of the main methods for quickly extracting cropland changes. However, existing methods were highly susceptible to seasonal differences due to the high heterogeneity of cultivated land. In this study, an integrated framework was proposed to perform change detection by incorporating phenological patterns and multi-source remote sensing images.There were two improvements in this proposed cropland change detection method: 1) the multi-source remote sensing images were utilized to fill the missing data within a time-series image stack by considering phenological patterns of cropland and 2) the Seq2Seq model considering phenological patterns was developed to extract changes in cropland directly. Compared to the traditional change detection methods, the proposed strategy was able to detect change process. Experiments demonstrated that the proposed method can significantly improved change detection accuracies, given a limited number of labeled samples.