COLLABORATIVE SYSTEM LABELING AND SATELLITE IMAGERY DATA PROCESSING FOR BUILDING CAPITAL STOCK ESTIMATION

• Main Author: Andri Astuti, Fitri
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/67488
• Institution: Institut Teknologi Bandung

The building capital stock can be an indicator in calculating the wealth of a region. In economics, the stock of building capital is calculated using the Perpetual Inventory Method (PIM). This method requires many input parameters such as the value of Gross Fixed Capital Formation (GFCF) based on current and constant prices, age of use, gross added value, and the price index of capital goods. The complexity of this method can be seen from the need for a backcasting method for all parameters and the distribution of retirement for each type of fixed assets so that the PIM method is good to use when the required parameters are available and the time required is long enough. Therefore, new data sources and calculation methods are needed to estimate building capital stock, namely using satellite imagery to detect building area automatically. Meanwhile, the development of cloud technology has accelerated the analysis of satellite imagery by various platforms, one of which is Google Earth Engine (GEE). However, the proposed solution has limitations in terms of the amount of labeling, collaboration, and memory for processing. Meanwhile, to make good data labeling requires good collaboration between users. A good labeling of satellite image data will produce a good machine learning model so that it can do a good classification. Based on this, this study presents the architecture of a collaborative labeling system and satellite imagery data processing for estimation of building capital stock. This study presents empirical results regarding usability tests, performance tests, and comparative analysis of the system. The results of the usability test are 93.03% which means the system is very feasible to use. Meanwhile, the result of the performance test is 0.999 which makes the system included in the 99% availability level criteria. The results of the comparative analysis show that the value of the capital stock of the proposed system output is statistically no difference with the PIM method with the retirement geometric distribution. Therefore, it can be concluded that the proposed system has succeeded in becoming one of the solutions in calculating the building capital stock from satellite imagery data.