إنتاج بيانات الكتل العمرانية بواسطة صور الأقمار الصناعية ذات الدقة المكانية العالية باستخدام نظم المعلومات الجغرافية والتعلم العميق

Abstract EN

1.

Study subject: the research presents a quantitative analysis of extracting the building footprints layer using one of the modern techniques through GIS software by modeling a set of procedures using artificial intelligence technology in extracting building layer data from high-resolution satellite images and deep learning models.

2.

objectives of the study: the study aims to reach the use of an integrated system using artificial intelligence technology and deep learning algorithms based on geographic information systems techniques to extract automated layer of buildings footprints by processing high-resolution satellite images in an attempt to develop one of the cartographic systems that can be used in producing thematic urban maps.

3.

study methodology: the methodology of this study depends on several sequential and interrelated practical steps that start with obtaining the inputs of the system.

first: obtaining the satellite visuals of the study area with high spatial accuracy, with some preliminary manipulations of the satellite visualization.

second: choosing the analysis model and extracting the buildings layer.

third: using the building footprints layer extraction model.

fourth: processing the results extracted from the system fifth: measuring the accuracy of the results.

4.

study datasets: this study relied on high-resolution satellite images with a resolution of 30 cm for the study area provided by digital globe co.

5.

results: the study reached a number of results, which can be divided as follows: A- results related to the methodology and model used: the study showed that how producing building footprints data can be done by satellite images with high resolution using geographic information systems and deep learning begins with defining the inputs and ensuring the standard accuracy of each input, starting from the selection of satellite images and their compatibility of the model that it used, which greatly reduces the time and effort required.

B- results related to the outputs of the model: • the study showed the importance of processing the outputs later, including generalizing building footprints, which affect the calculation of the accuracy of shape and area compared to manual digitizing.

• the results showed the effect of the regular and irregular building footprints pattern on the accuracy of the model in extracting the layer of buildings in the study area compared to the total expected real number, which reached 47.80% in al-Zahoor district, while it reached 11.63% in al-Dirah Distric.

• a similarity rate of 30% (threshold) was applied, and results with higher accuracy were obtained, while al-Zahoor and Dirah reached an accuracy of approximately 80% to 90%.

• two other areas were chosen to verify the results of the model's accuracy, which are the Hattin district in Riyadh and the al-Nahda district in Jeddah.

• the accuracy of the results also reached from 85% to 94%, as these results were confirmed after calculating accuracy, dice, Flscore, index, IoU, precision, and recall, by applying the python script to the output results through images of buildings footprint.

6.

conclusion: the study provided a detailed presentation on the methods of extracting the building footprints layer from high-resolution satellite images through the use of artificial intelligence and deep learning techniques in geographic information systems, where a set of algorithms were presented, which are the technical basis for machine learning, in terms of the mechanism of work and the accuracy of the results.