Before always wanted to develop a fully automated method of atmospheric correction, but the model-based method 6s, etc., requires a variety of aerosol parameters of the problem, leading to development has been at a standstill, recently saw MD, Professor of Nanjing University of Information Engineering, published in Remote Sensing the above correction methods atmosphere, the feeling is very interesting, and would like to achieve one.
The core idea of this approach is that atmospheric correction, only the sun sensor and the azimuth angle, elevation angle, the apparent reflectance data to, which is very convenient, without any input of atmospheric parameters, essentially, is in fact 6S correcting simplified model of the atmosphere only.
The following analysis of what its core principle:
If you have any questions, please contact qq1044625113; Email: [email protected].
I simply was implemented in pure matlab (reason useless python, because lazy), if the code needs to be in touch with me, here is the finished result of atmospheric correction:
From this result we can see, is still very good, surface reflectance basically in line with the visual effects, which I value were compared, the result is good!
This algorithm has the advantage that only azimuth, zenith parameters without the need of complex atmospheric parameters, etc., for the domestic number one high score, score II, high VI can be carried out, which is very convenient, it all can be applied to large-volume engineering production,
If using a parallel algorithm, the GPU computing, to a higher and higher efficiency level is not a problem, pure matlab code, i7 7700HQcpu After about 60 seconds executed, the IDL code needs more than 5 times longer. . . Why so slow IDL. . . Said an aside, I personally
Think the real development is to serve the high-volume engineering production, any algorithm castles in the air that makes no sense, which is MIT's idea is to have, and I after graduate school, changed the concept, whether it is because interest or What, I will think about,
If this thing is really useful, if you can really use them in large quantities, or that can change the existing mode. Let me give you an example, remote sensing images to the cloud, a lot of people write articles like to use images to replace the multi-phase cloud area, my heart really wanted to say, so sometimes only a
King, I go looking to replace it? So I think this study is actually of little significance, should be turned to how large quantities of automatic cloud masks, engineering studies automatic alternative, such as block GPU parallel and so on.
Well, first side-tracked here!