Using Artificial Intelligence for City Infrastructure Monitoring

Unauthorized development is an issue that impacts the safety and wellbeing of residents. For example, inadequate monitoring of the progress of building construction can lead to structures being located in high-risk areas such as those prone to floods.

There have been many cases where unauthorized buildings have been built on designated public open spaces under comprehensive architectural plans. These unauthorized developments can be discovered early on through regular monitoring of urban areas.

Statistics of demographics, commercial/manufacturing industries, and infrastructure are commonly used for urban development planning. However, due to the difficulty of retaining up-to-date information in rapidly growing cities, many developing and middle-income countries do not possess such high-quality statistical information. The lack of these statistics can prevent effective urban planning, and impede necessary services such as infrastructure development.

With high resolution satellite images analyzed using AI, cities can extract high-quality statistics that can be used for sustainable urban development.

Some examples of AI analysis are:

Road detection

Getting precise information on narrow roads can be hard because of shadows or dead angle of huge buildings and trees. AI can analyze road footprints from high resolution satellite images, and identify changes such as new construction and demolition by comparing satellite images from different time periods.

Building detection in dense urban areas

AI can detect unauthorized buildings such as informal settlements in rapidly growing cities. It can identify building changes from satellite images taken at different time periods.

Car detection in parking area/harbor facilities

Satellite images of cars in parking areas at commercial districts and harbor facilities can be analyzed by AI. Economic activities can be monitored indirectly by counting the number of cars in a designated area.

Paddy field detection

AI can analyze paddy fields. Blue lines show the extracted paddy field polygons in the images below. It can show agricultural land use and estimate yield of crops.

Paddy field detection using satellite data and AI. Image credits: Airbus DS and PASCO.

Land use classification

AI analysis can assist in classifying land use. Compared with the traditional image processing of maximum likelihood method, AI classification based on deep learning can classify land use more accurately by using the ground surface’s texture information from satellite images. It can also identify land use changes by comparing satellite images from two different time periods.

Land use classification by using satellite data and AI. Image credits: DigitalGlobe, Inc. and PASCO.

Estimate number of households

In this sample case, the number of households is estimated at a mesh size of 100 meters. After counting the number of buildings within the mesh, it is possible to estimate the number of households by multiplying the ratio of households to buildings in each municipality, city or town. Although this is a rough estimate, it is very useful in rapidly developing urban areas because urban planners have difficulties in collecting this information.

Number of households estimated using satellite data and AI. Image credits: DigitalGlobe, Inc. and PASCO.

Satellite images and AI can aid developing and middle-income countries in data collection services, which will allow for more efficient urban development planning.

The use of AI and satellite images is gaining momentum as the technology to deploy in putting together a regular collection of up-to-date statistics in urban areas. Specifically, satellite images can visualize infrastructure, such as roads and buildings, while AI can extract information from the images to create statistics, such as the location, number, and size of structures. Additionally, statistics, such as land use information, traffic volume via counting vehicles on the road, vegetation coverage, and building characteristics, can be automatically extracted from satellite images.

Satellite data is open data, which means it is publicly available. Governments can obtain the data from a provider (e.g. National Environmental Satellite Data and Information Service) or reseller in the market.

To develop an AI model and extract statistical information, it is necessary to (i) put the data infrastructure in place; (ii) collect satellite data; (iii) create training data for the AI model; and (iv) tune the AI model by using the training data.

Satellite providers and private companies also provide statistical information, which may be the fastest option for those lacking the capacity to develop an AI model.

Y. Nachmany and H. Alemohammad. 2019. Detecting Roads from Satellite Imagery in the Developing World. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshop.

United Nations Department of Economic and Social Affairs. 2019. The 2018 Revision of World Urbanization Prospects. New York: United Nations.