IRSAMap: High-Resolution Land Cover Vector Dataset

Data Source: Combining Jilin-1 satellite imagery (0.5 m), Google Earth, and the Open Cities AI Challenge, ensuring a uniform resolution of 0.5 m. Coverage: A total of 79 typical regions spanning six continents, with a total area of approximately 1,000 km². Among them, 16 regions in China (419 km²), 33 regions from Google Earth data (436 km²), and 30 regions from Open Cities AI data (155 km²). Data Scale: About 5,000 patches (1024×1024 pixels), totaling 1.8 million instances, covering 10 categories of natural and man-made features, including buildings, roads, water bodies, and vegetation. Sampling Strategy: Representative regions in China were selected according to nine major ecological zones, and regions abroad were divided by continent to ensure geographical diversity and balanced scenes. Method: Through an iterative annotation process involving manual labeling, model inference, and human correction, a global land cover vector dataset covering 79 regions was generated.

数据源：本数据集融合吉林一号（Jilin-1）卫星影像（0.5米分辨率）、谷歌地球（Google Earth）影像及开放城市AI挑战赛（Open Cities AI Challenge）数据，确保所有数据统一分辨率为0.5米。覆盖范围：数据集涵盖六大洲共79个典型区域，总面积约1000平方千米。其中包含中国境内区域16个（面积419平方千米）、源自谷歌地球的区域33个（面积436平方千米），以及源自开放城市AI挑战赛的区域30个（面积155平方千米）。数据规模：数据集包含约5000幅1024×1024像素的影像块，总计180万个标注实例，覆盖10类自然与人工地物，涵盖建筑物、道路、水体及植被等类别。采样策略：针对中国境内区域，依据九大生态分区选取代表性区域；境外区域则按大洲进行划分，以保障地理多样性与场景均衡性。数据制作方法：通过融合人工标注、模型推理与人工校验的迭代标注流程，生成了覆盖79个区域的全球土地覆盖矢量数据集。