Estimation of Economic Indicators using Residual Neural Network ResNet50

Spatially real and reliable of economic indicator is critical for both policy makers and researchers. The economic indicators are often used to reflect poverty in an area. At present, combining remote sensing imagery and machine learning are used as a new method for estimated economic indicator. In this paper, the deep residual network model ResNet50 is used to train remote sensing images to classify nighttime light intensity to obtains representative features, which are later used in ridge regression to fit to corresponding economic indicator. The Pearson coefficient between the predicted economic indicator and the actual economic indicator is calculated to evaluate the performance of the method. We estimated gross domestic product (GDP) and total retail sales of consumer goods (TRSCG) respectively, and the Pearson coefficients of GDP and TRSCG were both 0.85, which are higher than the results of linear regression model. The results show that the method has good prediction ability for GDP and TRSCG economic indicators. The satellite images in this paper are publicly available and therefore this method can be easily applied to other regions.