Artificial intelligence has estimated the degree of obesity of residents of certain areas of the United States using satellite photographs. To do this, the algorithm did not use images of individuals, but other data, such as the distribution of buildings and trees in the area. The results of the work are presented in the JAMA Network Open magazine.
Some public health problems are so great that they can be seen from space. In the new work, scientists used deep neural network learning to analyze satellite data from four residential areas in the United States. To do this, we used data on the urban environment, both natural and artificial: the presence of parks, the location of roads, pedestrian crossings, a variety of types of houses, and so on.
The source of the data was photographs of 1,695 neighborhoods in Los Angeles, Memphis, San Antonio and Seattle from the Google Maps service - in total about 150,000 images. From these images, the neural network extracted data on the distribution of vegetation, the position of roads and the presence of houses. Then another algorithm compared the information obtained with the obesity rate in the local population.
As a result, the creators of the neural network were able to estimate the number of obese people even more accurately than could be done based on the number of gyms and restaurants in the study area. They also managed to find a connection between planning parameters and per capita income.