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

Area of Science:

• Remote Sensing
• Computer Vision
• Agricultural Science

Background:

• Cropland delineation is crucial for agricultural surveys.
• Existing methods struggle with high-dimensional data from UAV RGB imagery.
• A gap exists in accurate cropland mapping using advanced aerial data.

Purpose of the Study:

• To develop a novel sparsity-based classification algorithm for cropland delineation.
• To address the limitations of current methods in processing UAV RGB data.
• To improve the accuracy and efficiency of agricultural plot identification.

Main Methods:

• Extracting multi-features from UAV RGB photographs.
• Designing a multi-feature association sparse model.
• Employing a breadth-first search to construct shape-adaptive windows.
• Iteratively solving an optimal objective to obtain the AMFSR algorithm.

Main Results:

• The proposed AMFSR algorithm achieved an overall accuracy (OA) of 92.3546%.
• The Kappa coefficient exceeded 0.8, indicating substantial agreement.
• Experimental results validated the model's generalization ability.

Conclusions:

• The AMFSR algorithm effectively delineates croplands using UAV RGB imagery.
• The sparsity-based approach overcomes challenges with high-dimensional data.
• This method offers a reliable tool for agricultural resource surveys and plot identification.