PM2.5 deep learning inversion method combining remote sensing data and social perception data

Abstract

The invention discloses a PM2.5 deep learning inversion method combining the remote sensing data and the social perception data. The PM2.5 deep learning inversion method comprises the steps of preprocessing the PM2.5 data of a ground station point, the remote sensing data, the social perception data and the auxiliary data; performing feature variable extraction and calculation on the multi-source data by using a geoscience space statistics and analysis method and a remote sensing information processing means; carrying out space-time matching on the multi-source data in a gridding mode, taking a grid with a ground station point true value as a training sample, and generating a multi-source data set with uniform space and time; normalizing the grid data set with the site PM2.5 true value, then inputting the grid data set into a deep learning model for training, and inverting the unknown grid PM2.5 concentration through the model after the grid data set passes the verification; and carrying out fine PM2.5 space-time distribution mapping on an inversion result. According to the method, the multi-source information can be effectively mined by using a deep learning technology, the defect of a traditional statistical model in a nonlinear problem is overcome, and the higher inversion precision and the finer space-time PM2.5 distribution are obtained.

Description

technical field [0001] The invention belongs to the field of remote sensing image processing and information application, and relates to a method for obtaining PM2.5 concentration, in particular to a method for retrieving fine spatiotemporal PM2.5 concentration based on deep learning combined with remote sensing data and social perception data. Background technique [0002] Fine spatio-temporal PM2.5 concentration distribution plays an important role in environmental monitoring and health assessment applications. The distribution of existing stations is sparse and uneven, and station observations cannot meet the application requirements. Obtaining PM2.5 temporal and spatial continuous distribution data has attracted wide attention. The causes of PM2.5 are complex and change rapidly, and its concentration is affected by natural and human factors at the same time, and there is a complex nonlinear relationship between each factor and PM2.5. Therefore, how to effectively combin...

AI Technical Summary

Problems solved by technology

[0003] At present, PM2.5 concentration inversion methods mainly include physical and chemical model simulation method and statistical model method. The former relies on the input of many model parameters and is complicated to realize; while the technology of statistical model method to estimate PM2.5 concentration relies on its high precision and easy implementation. The advantages are widely used

There are two main deficiencies in the exis

Images