Creation of noise processing code by wavlet transformation of satellite image data time series in CUDA/C++ (mother: coiflet1-5 etc.)
Budget: $30 – $250 USD
I would like you to create a code to denoise the pixel value data (vegetation index NDVI) of the xyz 3D data (x longitude, y latitude, z time) for each x,y pixel in the z time direction.
The denoising should be done with wavlet (mother wavelet should be coifle1-5 (preferably symlet (4-15) and daubechies (2-24order) versions), and I need a C++ rewrite of the fortran namelist in my reference file. (I would like to use the coiflet1~5 freely depending on the input).
The only language that can be used is cuda, and we would like you to create x,y,z parallel or two types of x,y parallel code, using the following website as a reference.
Also, we would like you to use this namelist to set the xyz number of images and other parameters for coiflet3-5, and to input all the data. (Please do not use C macros to set values). If possible, all arrays to be handled should be newelist. If possible, please treat all arrays dynamically with new/free and do not use static arrays (small arrays in the kernel are acceptable, but please make them as dynamic as possible).
https://pbcglab.jp/cgi-bin/wiki/index.php?CUDA%E3%81%A7%E9%80%A3%E7%B6%9A%E3%82%A6%E3%82%A7%E3%83%BC%E3%83%96%E3%83%AC%E3%83%83%E3 %83%88%E5%A4%89%E6%8F%9B
If you respond, I will send you my C++ version of the namelist file and sample satellite data.
For testing, I would like you to download my sample data or the MODIS satellite image of the Mekong Delta area of Vietnam from NASA's hp, etc. and test it with the NDVI/NDWI 16-day composite and 8-day composite data. # i.e., if you can represent 2- and 3-fold crop peaks in the farmland pixels, it is OK.
I will also send you the data after processing with another noise removal signal processing program I have prepared. If the product of the code removed by this other program and the position of the maxima peaks (convex top vertices) in the time direction coincide to some extent in the spatio-temporal direction, I will consider it complete.
The denoising should be done with wavlet (mother wavelet should be coifle1-5 (preferably symlet (4-15) and daubechies (2-24order) versions), and I need a C++ rewrite of the fortran namelist in my reference file. (I would like to use the coiflet1~5 freely depending on the input).
The only language that can be used is cuda, and we would like you to create x,y,z parallel or two types of x,y parallel code, using the following website as a reference.
Also, we would like you to use this namelist to set the xyz number of images and other parameters for coiflet3-5, and to input all the data. (Please do not use C macros to set values). If possible, all arrays to be handled should be newelist. If possible, please treat all arrays dynamically with new/free and do not use static arrays (small arrays in the kernel are acceptable, but please make them as dynamic as possible).
https://pbcglab.jp/cgi-bin/wiki/index.php?CUDA%E3%81%A7%E9%80%A3%E7%B6%9A%E3%82%A6%E3%82%A7%E3%83%BC%E3%83%96%E3%83%AC%E3%83%83%E3 %83%88%E5%A4%89%E6%8F%9B
If you respond, I will send you my C++ version of the namelist file and sample satellite data.
For testing, I would like you to download my sample data or the MODIS satellite image of the Mekong Delta area of Vietnam from NASA's hp, etc. and test it with the NDVI/NDWI 16-day composite and 8-day composite data. # i.e., if you can represent 2- and 3-fold crop peaks in the farmland pixels, it is OK.
I will also send you the data after processing with another noise removal signal processing program I have prepared. If the product of the code removed by this other program and the position of the maxima peaks (convex top vertices) in the time direction coincide to some extent in the spatio-temporal direction, I will consider it complete.