Semantics-based high resolution reconstruction method of nighttime light remote sensing image

A semantics-based high resolution reconstruction method of a nighttime light remote sensing image, the method comprising:

• 1) constructing a sample data set;
• wherein the sample data set comprises a plurality of data groups, and each data group comprises an NTL remote sensing image with low-spatial-resolution (LR NTL image), and an NTL remote sensing image with high-spatial-resolution (HR NTL image) and light semantics information consistent in spatial position with the LR NTL image;
• 2) constructing a reconstruction model, wherein a network architecture model of the reconstruction model adopts Unet;
• 3) performing training and validation on the reconstruction model by using the sample data set to obtain an optimized reconstruction model;
• when the reconstruction model is trained, inputting the LR NTL image and the light semantics information consistent in the spatial position with the LR NTL image to the reconstruction model which outputs a reconstructed HR NTL image;
• comparing the reconstructed HR NTL image with the HR NTL image consistent in the spatial position with the LR NTL image in the sample data set to measure an error therebetween by using a loss function, and an optimal Unet reconstruction model is obtained when the loss function is minimized;
• and
• 4) taking a to-be-reconstructed LR NTL image and light semantics information corresponding to the to-be-reconstructed LR NTL image as an input of the optimized reconstruction model, and outputting, by the optimized Unet reconstruction model in 3), a HR NTL image obtained through resolution reconstruction.