The Global Context-Enhanced UPerNet for Cropland Extraction

Cropland is a key resource to ensure food security and promote agricultural development. Studying how to quickly and accurately extract cropland from remote sensing images is of great significance. To solve the problems of insufficient utilization of multi-scale features and weak global perception in existing cropland extraction methods, this paper proposes an improved UPerNet model-GCUperNet. By introducing the Global Context Module (GC Module), the model’s perception of long-distance dependencies and complex scenes in remote sensing images is effectively enhanced. The GC Module can capture the global semantic information of remote sensing images based on the fusion of multi-scale features, and improve the extraction of cropland boundaries and detail recognition. On our self-made Sichuan Yibin cropland extraction dataset, GCUperNet has increased the indicators of Accuracy, IoU, and F1 score by 1.18%, 1.72%, and 0.56% respectively, which proves its potential in the task of cropland extraction from remote sensing images.