Method for extracting evergreen vegetation in urban areas based on high-resolution remote sensing images

The extraction of evergreen vegetation in urban areas holds significant importance for environmental monitoring and sustainable urban development. To address the limitations of existing visible light vegetation indices in environmental adaptability and the critical role of sample annotation in vegetation segmentation, this paper proposes a sample optimization method that integrates color theory and EfficientSAM, aiming to enhance the accuracy of evergreen vegetation extraction in urban areas. This method utilizes high-resolution remote sensing images from the visible light spectrum, combining the color sensitivity of visible light vegetation indices with the prompting capabilities of EfficientSAM to optimize samples. The optimized results are then used to train semantic segmentation models, effectively achieving precise extraction of evergreen vegetation. Experimental results demonstrate that the proposed method achieves improvements of 83.83%, 92.23%, 89.72%, and 90.96% in mI, mP, mR, and mF metrics, respectively, compared to traditional manually annotated sample training results. Furthermore, the method effectively distinguishes vegetation in water bodies from evergreen vegetation, providing a valuable reference for the accurate extraction of evergreen vegetation using visible-band imagery.