Slum Area Detection in Dhaka City, Bangladesh, Using Satellite Remote Sensing and Deep Learning

Dhaka City, a rapidly growing megacity and high-density developing area in Bangladesh, faces numerous urbanization challenges. Among these, informal settlements or slum areas pose serious concerns for authorities and policymakers striving to improve living conditions for slum dwellers. Dhaka City’s slums are characterized by high-density, one- or two-story structures made of low-cost materials, often located adjacent to natural water bodies and lacking basic public infrastructure (Ahmed 2014; Iqbal 2014; Jahan 2012). The need for accurate and up-to-date data on these settlements is crucial for researchers and policymakers, particularly for implementing slum upgradation and resettlement programs. However, traditional field surveys for mapping slum areas are timeconsuming and require significant financial resources. This research explores the feasibility of satellite remote sensing-based slum area detection as an alternative to physical surveys by evaluating the accuracy of slum area mapping using satellite imagery at different resolutions. Specifically, it compares the detection performance of PlanetScope (3m) and Maxar (0.3m) imagery using the U-Net deep learning architecture and U-Net with attention mechanisms. While similar methods applied in other developing cities have shown promising results, the findings of this study reveal that slum area detection accuracy is almost the same for the two different image resolutions, ranging from 72-79 percent overall accuracy (F1 score). This indicates the challenges associated with automated slum area mapping but also highlights the potential of high-resolution satellite imagery for future studies. The study also highlights the importance of training data collection for deep learning-based applications. Ultimately, this research contributes to addressing urban challenges such as unplanned housing, lack of infrastructure, and inaccessibility of essential services by supporting data-driven policy decisions for slum upgradation and resettlement initiatives.