<![CDATA[Urbanization in West Jakarta was recorded as the highest and most stable among Jakarta’s administrative regions in 2026, which demands the availability of accurate and up-to-date ground-truth spatial mapping data. In response to this demand, the utilization of Big Data in the form of very high-resolution satellite imagery enables the provision of information at a granular level, which can be translated into spatial planning maps through an appropriate modeling approach. In the case of spatial mapping in the West Jakarta region, this study examines the application of a deep learning-based semantic segmentation method using the UNet++ architecture with an InceptionResNetV2 encoder to extract building, road, and other object classes from 0.5-meter spatial resolution Pleiades satellite imagery. In developing this deep learning model, the model was developed and evaluated through an iterative methodological experiment, encompassing CLAHE, label cleaning, Focal Loss, an SCSE module, gamma correction, and an overlapping-tiling patch scheme. The results show an improvement in mIoU from 0.5789 (baseline model) to 0.7145 during the tuning stage, reaching 0.7352 with an F1-score of 0.8467 during the evaluation stage using the Patch 256 + Stride 128 configuration. The resulting map demonstrates potential in supporting spatial planning optimization in rapidly developing urban areas such as West Jakarta.]]>
Copyrights © 2026
