Abstract
As urban populations have grown in the last century, so has the rate of obesity, diabetes, and high blood pressure in those populations. For government and health agencies, getting accurate and up-to-date data on the rate of occurrence of these diseases within neighborhoods, census tracts, and larger urban regions is critical for care but cost prohibitive. In parallel to these developments, a growing body of research has shown significant correlations between the built environment and health. Therefore, the development of models able to estimate health measures from the characteristics of the built environment could significantly impact the management of these diseases and inform urban planning. This research addresses this problem through the development of an estimation model that uses deep learning and satellite imagery to estimate the rate of diabetes, high blood pressure, and people overweight in US census tracts.
Figure 1 a) Shows the activation maps of the CNN after the first convolutional block for a census tract with a high rate of diabetes. b) Shows the activation maps of the CNN after the second convolutional block with a high rate of diabetes. c) Shows the most active activation map in the first convolutional block. d) Shows the most active activation map in the second convolutional block.