Identification of Object-Based Residential Density Using the Mask R-CNN Method

This research employs the Mask R-CNN algorithm to identify and categorize residential density in Makassar City utilizing satellite imagery. This algorithm allows for pixel-based object segmentation with high precision so that residential areas can be grouped into three density categories: high, medium, and low. The satellite imagery dataset was obtained from Google Earth, manually annotated using polygons, and trained using Mask R-CNN with the ResNet-50 backbone. The evaluation means Average Precision (mAP), loss graphs, and confusion matrices. The results showed an overall accuracy of93%, with an accuracy of90% for the high-density category and 80% for the medium and low categories, respectively. This algorithm provides an in-depth understanding of urban density patterns and spatial relationships between structures, thus supporting more effective urban planning. These findings are expected to help the government formulate data-based development policies to overcome population density problems, especially in urban areas while improving the community's quality of life through sustainable regional management.