I have this classification problem , classifying crops using satellite imagery , I have two kinds of data optical ( 20 features )and radar data ( 6 features ) . There are two types of satellites each for our two data categories , they takes several images all year long ( not in a regular interval ) thus the temporal aspect of our data and the usage of recurrent networks. There’s a common issue with optical imagery concerning blockage , if there clouds are and most of cases there are , we use a method called gap-filling ( interpolate the missing data points from previous dates ) but it leads too many repetition in the data . on the other hand , Radar data they don’t have this issue but they give , alone ( 6 features ) or with optical data ( 20 + 6 features ) poor classification Overall Accuracy . So my question goes like this , is there a way of using both data but dynamically modifying weights in the neural nets so it becomes more sensitive to these kind of problems ( or with more examples the neural net will naturally become more sensitive to the problem ) , Sorry for the long post I just wanted to explain the problem at hand fully.
And thank you.