Automatic Urban Green Space Identification Based on GEE and Sentinel-2 Imagery

In contemporary urban ecological development and resource management, the establishment of urban green spaces is a crucial indicator of urban ecological quality and plays a significant role in improving the ecological environment. Therefore, accurate identification of green spaces is vital for urban ecological development. Remote sensing technology is commonly used for green space identification; however, traditional methods for sample point labeling are cumbersome and labor-intensive, limiting the application of large-scale data processing. Hence, a more efficient method for automatic sample point generation is needed. To alleviate the burden of manual sample point labeling, this study proposes an algorithm and selection rules for automatic sample point generation based on the Google Earth Engine (GEE) cloud computing platform. The algorithm relies on adding random points within the pixel mask layer of an unsupervised classification result map. By merging closely spaced random points, the algorithm purifies and assigns attribute values to generate sample points with attributes. The sample points generated by this method are comparable to those manually labeled on maps. This algorithm and selection rules not only apply to the specific context of this study but also have considerable potential for selecting sample points in other land cover types. Although the classification result accuracy (91.9%) is slightly lower compared to manual labeling (98.7%), this method significantly enhances work efficiency and reduces the labor required for manual labeling. Additionally, the algorithm demonstrates good portability. The automatic sample point generation algorithm, combined with the sample point selection rules, has significant application prospects in map data processing and can provide strong support for urban ecological development.