University of Twente Student Theses
Deep learning for integration of Satellite images and Google Streetview for mapping informal settlements (slums)
Wang, Pengyu (2024) Deep learning for integration of Satellite images and Google Streetview for mapping informal settlements (slums).
Full text not available from this repository.
| Full Text Status: | Access to this publication is restricted |
| Abstract: | Over the past two decades, rapid urbanization has led to a significant increase in informal settlements, particularly in developing regions. These informal settlements present substantial challenges for urban planning, public health, and sustainable development. However, the reliance on commercial data and inefficiencies in multi-source data fusion in previous studies have hindered the feasibility of informal settlements mapping efforts in resource-constrained areas. This study aims to develop an innovative neural network architecture to extract and fuse features from two free-cost datasets—Sentinel-2 satellite data and Google Street View imagery—to improve the computational efficiency and accuracy of informal settlement classification. Additionally, a dual-branch approach was employed to compare and evaluate the classification metrics of informal settlements using the Sentinel-2 only model against the feature fusion model. The results indicate that the integration of street view data with Sentinel- 2 satellite data significantly improves the classification accuracy for informal settlements. The precision for the "slum" category increased from 59% to 68%, enhancing the model's ability to accurately identify informal settlements while reducing the false positive rate. The recall for slums also improved from 63% to 65%, indicating better detection of actual informal settlements. The F1 score for slums increased from 61% to 66%, reflecting a more balanced and reliable classification outcome through multi-source data integration. Despite these promising results, the improvement in performance metrics suggests that there remains significant room for enhancing the fusion model's predictive capability for informal settlements. Future research could explore incorporating more spectral bands of varying resolutions and merging the dual-branch model into a single architecture to achieve even better results. |
| Item Type: | Essay (Master) |
| Faculty: | ITC: Faculty of Geo-information Science and Earth Observation |
| Subject: | 74 (human) geography, cartography, town and country planning, demography |
| Programme: | Geoinformation Science and Earth Observation MSc (75014) |
| Link to this item: | https://purl.utwente.nl/essays/103530 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page