MLC-BD: Multi-Platform Land Cover Imagery Dataset of Bangladesh

MLC-BD is a large-scale, multi-source satellite image dataset containing 10,683 manually annotated PNG images representing seven major land-cover categories across Bangladesh. The dataset was compiled from three widely used web-based mapping platforms: ArcGIS Map (3,556 images), Bing Maps (3,632 images), and Google Maps (3,495 images). Images were collected through systematic visual exploration of satellite and aerial basemap imagery covering diverse geographic regions of Bangladesh and were manually assigned to one of seven land-cover classes: Agricultural Land, Beaches/Coastline, Built-up Areas, Grasslands/Open Land, Hills/Mountains, Trees/Forests, and Water Bodies. Each image subsequently underwent quality inspection, standardized labeling, and structured curation to ensure dataset consistency. The dataset is well balanced, with 1,509 to 1,547 images per class, and contains images with varying spatial dimensions ranging from 88 to 506 pixels in width and 72 to 540 pixels in height. MLC-BD is designed to support research in computer vision, remote sensing, and geospatial artificial intelligence, including land-cover recognition, transfer learning, cross-source domain generalization, and model robustness evaluation across heterogeneous imagery providers. By focusing exclusively on Bangladesh, a geographically diverse and densely populated South Asian deltaic nation that remains underrepresented in publicly available remote sensing benchmarks, MLC-BD provides a valuable resource for developing and evaluating vision-based geospatial analysis methods in real-world, multi-source environments.MLC-BD是一款大规模多源卫星图像数据集，包含10683张经人工标注的PNG图像，覆盖孟加拉国境内7大类主要土地覆盖类型。该数据集采集自三个广泛使用的网络地图平台：ArcGIS地图（3556张图像）、必应地图（3632张图像）以及谷歌地图（3495张图像）。研究人员通过对覆盖孟加拉国多样地理区域的卫星与航空底图影像开展系统性目视探查完成图像采集，并将每张图像手动归类至7类土地覆盖类别之一：农用地、海滩/海岸线、建成区、草地/开阔地、丘陵/山地、树木/森林以及水体。后续针对所有图像开展质量检验、标准化标注与结构化整理工作，以保障数据集的一致性。该数据集类别分布均衡，每类图像的数量介于1509至1547张之间，图像空间尺寸跨度广泛，宽度范围为88至506像素，高度范围为72至540像素。MLC-BD旨在为计算机视觉、遥感以及地理空间人工智能（Geospatial Artificial Intelligence）领域的研究提供支撑，涵盖土地覆盖识别、迁移学习、跨源域泛化以及针对异构影像提供商的模型鲁棒性评估等研究方向。由于仅聚焦孟加拉国——一个地理类型多样、人口稠密的南亚三角洲国家，且在公开可用的遥感基准数据集中长期代表性不足——MLC-BD为在真实多源环境下开发与评估基于视觉的地理空间分析方法提供了极具价值的研究资源。