Multi-feature sparse representation based on adaptive graph constraint for cropland delineation

Abstract

Cropland delineation is the basis of agricultural resource surveys and many algorithms for plot identification have been studied. However, there is still a vacancy in SRC for cropland delineation with the high-dimensional data extracted from UAV RGB photographs. In order to address this problem, a new sparsity-based classification algorithm is proposed. Firstly, the multi-feature association sparse model is designed by extracting the multi-feature of UAV RGB photographs. Next, the samples with similar characteristics are hunted with the breadth-first principle to construct a shape-adaptive window for each test. Finally, an algorithm, multi-feature sparse representation based on adaptive graph constraint (AMFSR), is obtained by solving the optimal objective iteratively. Experimental results show that the overall accuracy (OA) of AMFSR reaches 92.3546% and the Kappa is greater than 0.8. Furthermore, experiments have demonstrated that the model also has a generalization ability.